the quarterly journal of economics · 2018. 7. 11. · the quarterly journal of economics vol. 133...

THE

QUARTERLY JOURNALOF ECONOMICS

Vol 133 2018 Issue 3

THE IMPACTS OF NEIGHBORHOODS ONINTERGENERATIONAL MOBILITY I CHILDHOOD

EXPOSURE EFFECTSlowast

RAJ CHETTY AND NATHANIEL HENDREN

We show that the neighborhoods in which children grow up shape their earn-ings college attendance rates and fertility and marriage patterns by studyingmore than 7 million families who move across commuting zones and counties inthe United States Exploiting variation in the age of children when families movewe find that neighborhoods have significant childhood exposure effects the out-comes of children whose families move to a better neighborhoodmdashas measured bythe outcomes of children already living theremdashimprove linearly in proportion tothe amount of time they spend growing up in that area at a rate of approximately4 per year of exposure We distinguish the causal effects of neighborhoods from

lowastAn earlier version of this article was circulated as Part I of ldquoThe Impactsof Neighborhoods on Intergenerational Mobility Childhood Exposure Effects andCounty Level Estimatesrdquo The opinions expressed in this article are those of theauthors alone and do not necessarily reflect the views of the Internal RevenueService or the US Treasury Department This work is a component of a largerproject examining the effects of tax expenditures on the budget deficit and eco-nomic activity All results based on tax data in this article are constructed usingstatistics originally reported in the SOI Working Paper ldquoThe Economic Impactsof Tax Expenditures Evidence from Spatial Variation across the USrdquo approvedunder IRS contract TIRNO-12-P-00374 We thank Gary Chamberlain Maximil-ian Kasy Lawrence Katz Jesse Shapiro and numerous seminar participants forhelpful comments and discussions Sarah Abraham Alex Bell Augustin Berg-eron Michael Droste Niklas Flamang Jamie Fogel Robert Fluegge NikolausHildebrand Alex Olssen Jordan Richmond Benjamin Scuderi Priyanka Shendeand our other predoctoral fellows provided outstanding research assistance Thisresearch was funded by the National Science Foundation the Lab for EconomicApplications and Policy at Harvard Stanford University and the Laura and JohnArnold Foundation

Ccopy The Author(s) 2018 Published by Oxford University Press on behalf of the Presi-dent and Fellows of Harvard College All rights reserved For Permissions please emailjournalspermissionsoupcomThe Quarterly Journal of Economics (2018) 1107ndash1162 doi101093qjeqjy007Advance Access publication on February 10 2018

1107Downloaded from httpsacademicoupcomqjearticle-abstract133311074850660by Harvard Library useron 11 July 2018

1108 THE QUARTERLY JOURNAL OF ECONOMICS

confounding factors by comparing the outcomes of siblings within families study-ing moves triggered by displacement shocks and exploiting sharp variation inpredicted place effects across birth cohorts genders and quantiles to implementoveridentification tests The findings show that neighborhoods affect intergenera-tional mobility primarily through childhood exposure helping reconcile conflictingresults in the prior literature JEL Codes J62 C00 R00

I INTRODUCTION

To what extent are childrenrsquos economic opportunities shapedby the neighborhoods in which they grow up Despite extensiveresearch the answer to this question remains debated Observa-tional studies have documented significant variation across neigh-borhoods in economic outcomes (eg Wilson 1987 Jencks andMayer 1990 Massey and Denton 1993 Sampson Morenoff andGannon-Rowley 2002 Sharkey and Faber 2014) However exper-imental studies of families that move have traditionally foundlittle evidence that neighborhoods affect economic outcomes (egKatz Kling and Liebman 2001 Oreopoulos 2003 Ludwig et al2013)

Using deidentified tax records covering the US populationwe present new quasi-experimental evidence on the effects ofneighborhoods on intergenerational mobility that reconcile theconflicting findings of prior work and shed light on the mecha-nisms through which neighborhoods affect childrenrsquos outcomesOur analysis consists of two articles In this article we measurethe degree to which the differences in intergenerational mobilityacross areas in observational data are driven by causal effects ofplace In the second article (Chetty and Hendren 2018a) we buildon the research design developed here to construct estimates ofthe causal effect of growing up in each county in the United Stateson childrenrsquos long-term outcomes and characterize the features ofareas that produce good outcomes

Our analysis is motivated by our previous work showing thatchildrenrsquos expected incomes conditional on their parentsrsquo incomesvary substantially with the area (commuting zone or county)in which they grow up (Chetty et al 2014)1 This geographic

1 To maximize statistical precision we characterize neighborhood (or ldquoplacerdquo)effects at two broad geographies counties and commuting zones (CZs) whichare aggregations of counties that are similar to metro areas but cover the entireUnited States including rural areas Counties are much larger than the typical

Downloaded from httpsacademicoupcomqjearticle-abstract133311074850660by Harvard Library useron 11 July 2018

NEIGHBORHOODS AND INTERGENERATIONAL MOBILITY I 1109

variation in intergenerational mobility could be driven by twovery different sources One possibility is that neighborhoods havecausal effects on economic mobility that is moving a given childto a different neighborhood would change his or her life outcomesAnother possibility is that the observed geographic variation isdue to systematic differences in the types of people living in eacharea such as differences in demographics or wealth

We assess the relative importance of these two explanationsby asking whether children who move to areas with higher ratesof upward income mobility among ldquopermanent residentsrdquo havebetter outcomes themselves2 Because moving is an endogenouschoice simple comparisons of the outcomes of children whose fam-ilies move to different areas confound causal effects of place withselection effects (differences in unobservables) We address thisidentification problem by exploiting variation in the timing of chil-drenrsquos moves across areas3 We compare the outcomes of childrenwho moved to a better (or worse) area at different ages to identifythe rate at which the outcomes of children who move converge tothose of the permanent residents4 The identification assumptionunderlying our research design is that the selection effects (chil-drenrsquos unobservables) associated with moving to a better versusworse area do not vary with the age of the child when the fam-ily moves This is a strong assumption one that could plausiblybe violated for several reasons For instance families who moveto better areas when their children are young may be more edu-cated or invest more in their children in other ways We presentevidence supporting the validity of this identification assumptionafter presenting a set of baseline results

geographic units used to define ldquoneighborhoodsrdquo however the variance of placeeffects across the broad geographies we study is a lower bound for the total varianceof neighborhood effects which would include additional local variation

2 We define ldquopermanent residentsrdquo as the parents who stay in the same CZ (orin the county-level analysis the same county) throughout the period we observe(1996ndash2012)

3 Several recent studies have used movers to identify causal effects of placeson other outcomes using event-study designs comparing individualsrsquo outcomesbefore versus after they move (eg Chetty Friedman and Saez 2013 FinkelsteinGentzkow and Williams 2016) We use a different research design because wenaturally do not have premove data on income in adulthood when studying theimpact of moving during childhood

4 Throughout the article we refer to areas where children have better out-comes in adulthood as ldquobetterrdquo neighborhoods We use this terminology withoutany normative connotation as there are of course many other amenities of neigh-borhoods that may be relevant from a normative perspective



In our baseline analysis we focus on families with childrenborn between 1980 and 1988 who moved once across commut-ing zones (CZs) between 1997 and 2010 We find that on aver-age spending an additional year in a CZ where the mean incomerank of children of permanent residents is 1 percentile higher(at a given level of parental income) increases a childrsquos incomerank in adulthood by approximately 004 percentiles That is theincomes of children who move converge to the incomes of per-manent residents in the destination at a rate of 4 per year ofchildhood exposure Symmetrically moving to an area where per-manent residents have worse incomes reduces a childrsquos expectedincome by 4 per year When analyzing children who move morethan once during childhood we find that childrenrsquos incomes varyin proportion to the amount of time they spend in an area ratherthan the specific ages during which they live in that area aswould be the case in a model of ldquocritical age effectsrdquo (eg Lynchand Smith 2005)

Together these results imply that neighborhoods have sub-stantial childhood exposure effects every additional year of child-hood spent in a better environment improves a childrsquos long-termoutcomes The outcomes of children who move converge linearly tothe outcomes of permanent residents in the destination over theage range we are able to study in our data (ages 9 to 23) Henceannual exposure effects are approximately constant moving toa better area at age 9 instead of 10 is associated with the sameincrease in income as moving to that area at age 15 instead of16 The exposure effects persist until children are in their earlytwenties We find similar childhood exposure effects for severalother outcomes including rates of college attendance marriageand teenage birth We also find similar exposure effects whenfamilies move across counties

Our estimates imply that the majority of the observed vari-ation in outcomes across areas is due to causal effects of placeThe convergence rate of 4 per year of exposure between the agesof 9 and 23 implies that children who move at age 9 would pickup about (23 minus 9) times 4 = 56 of the observed difference in per-manent residentsrsquo outcomes between their origin and destinationCZs If we extrapolate to earlier ages by assuming that the rate ofconvergence remains at 4 even before age 9mdasha strong assump-tion that should be evaluated in future workmdashour estimates implythat children who move at birth to a better area and stay there for



20 years would pick up about 80 of the difference in permanentresidentsrsquo outcomes between their origins and destinations

As noted the identification assumption underlying the inter-pretation of the 4 convergence rate as a causal exposure effect isthat the potential outcomes of children who move to better versusworse areas do not vary with the age at which they move We usefour approaches to evaluate this assumption controlling for ob-servable fixed family characteristics controlling for time-varyingobservable characteristics isolating plausibly exogenous movestriggered by aggregate displacement shocks and implementing aset of outcome-based placebo tests The first three approaches arefamiliar techniques in the treatment effects literature whereasthe fourth exploits the multi-dimensional nature of the treatmentswe study and the precision afforded by our large samples to im-plement overidentification tests of the exposure effect model

To implement the first approach we begin by controlling forfactors that are fixed within the family (eg parent education) byincluding family fixed effects5 This approach identifies exposureeffects from comparisons between siblings by asking whether thedifference in earnings outcomes between two siblings who move toa new area is proportional to their age difference interacted withpermanent residentsrsquo outcomes in the destination We estimate anannual exposure effect of approximately 4 per year with familyfixed effects very similar to our baseline estimate

These sibling comparisons address confounds due to factorsthat are fixed within families but they do not account for time-varying factors such as a change in family environment at thetime of the move that directly affects children in proportion toexposure time independent of neighborhoods We cannot observe

5 The idea of using sibling comparisons to better isolate neighborhood effectsdates to the seminal review by Jencks and Mayer (1990) Plotnick and Hoffman(1996) and Aaronson (1998) implement this idea using data on 742 sibling pairsfrom the Panel Study of Income Dynamics but reach conflicting conclusions dueto differences in sample and econometric specifications Several studies also usesibling comparisons to identify critical periods that shape immigrantsrsquo long-termoutcomes (Basu 2010 van den Berg et al 2014) Our approach differs from thesestudies in that we focus on how the difference in siblingsrsquo outcomes covaries withthe outcomes of permanent residents in the destination neighborhood whereasthe studies of immigrants estimate the mean difference in siblingsrsquo outcomes asa function of their age gap This allows us to separate the role of neighborhoodexposure from changes within the family that also generate exposure-dependentdifferences across siblings such as changes in income or wealth when a familymoves to a new country



all such time-varying factors but we do observe two particularlyimportant characteristics of the family environment in each yearincome and marital status In our second approach we show thatcontrolling flexibly for changes in income and marital status in-teracted with the age of the child at the time of the move has noimpact on the exposure effect estimates

The preceding results rule out confounds due to observ-able factors such as income but they do not address potentialconfounds due to unobservable factors In particular whateverevent endogenously induced a family to move (eg a wealth shock)could also have had direct effects on their childrenrsquos outcomes Ourthird approach addresses the problem of bias associated with en-dogenous choice by focusing on a subset of moves that are morelikely to be driven by exogenous aggregate shocks In particularwe identify moves that occur as part of large outflows from ZIPcodes often caused by natural disasters or local plant closures Wereplicate our baseline design within this subsample of displacedmovers comparing the outcomes of children who move to differentdestinations at different ages We obtain similar exposure effectestimates for displaced households mitigating concerns that ourbaseline estimates are biased by omitted variables correlated witha householdrsquos choice of when to move6

Although the evidence from the first three approachesstrongly supports the validity of the identification assumptioneach approach itself rests on assumptions (selection on observ-ables and exogeneity of the displacement shocks) that could po-tentially be violated We therefore turn to a fourth approach aset of placebo (overidentification) tests that exploit heterogene-ity in permanent residentsrsquo outcomes across subgroups which inour view provides the most compelling method of assessing thevalidity of the research design

We begin by analyzing heterogeneity across birth cohortsAlthough outcomes within CZs are highly persistent over timesome places improve and others decline Exploiting this varia-tion we find using multivariable regressions that the outcomesof children who move to a new area converge to the outcomes ofpermanent residents of the destination in their own birth cohortbut are unrelated to those of the preceding and subsequent birth

6 We eliminate variation due to individualsrsquo endogenous choices of whereto move in these specifications by instrumenting for each householdrsquos change inneighborhood quality using the average change in neighborhood quality of thosewho move out of the ZIP code during the years in our sample



cohorts (conditional on their own birth cohortrsquos predictions) Suchcohort-specific convergence is precisely what one would expect ifplaces have causal effects on childrenrsquos outcomes in proportion toexposure time but would be unlikely to emerge from sorting orother omitted variables because the cohort-specific effects are onlyrealized with a long time lag after children grow up

We implement analogous placebo tests by exploiting varia-tion in the distribution of outcomes across areas For instancelow-income children who spend their entire childhood in Bostonand San Francisco have similar incomes on average in adulthoodbut children in San Francisco are more likely to end up in the up-per or lower tail of the income distribution (ie either in the top10 or not employed) The causal exposure effects model predictsconvergence not just at the mean but across the entire distribu-tion in contrast it would be unlikely that omitted variables (suchas changes in parent wealth) would happen to perfectly replicatethe entire distribution of outcomes in each area in proportion toexposure time In practice we find quantile-specific convergencecontrolling for mean incomes childrenrsquos incomes converge to pre-dicted incomes in the destination across the distribution in pro-portion to exposure time at a rate of about 4 a year

Finally we implement placebo tests exploiting heterogene-ity in permanent residentsrsquo outcomes across genders Althoughearnings outcomes are highly correlated across genders there aresome places where boys do worse than girls (eg areas with con-centrated poverty) and vice versa When a family with a daughterand a son moves to an area that is especially good for boys theirson does better than their daughter in proportion to the numberof years they spend in the new area Once again if our findings ofneighborhood exposure effects were driven by sorting or omittedvariables one would not expect to find such gender-specific con-vergence in incomes unless families are fully aware of the exactgender differences in incomes across areas and sort to neighbor-hoods based on these gender differences

In sum the four tests we implement imply that any omit-ted variable that generates bias in our exposure effect estimatesmust (i) operate within families in proportion to exposure time(ii) be orthogonal to changes in parental income and marital sta-tus (iii) persist in the presence of moves induced by displacementshocks and (iv) precisely replicate permanent residentsrsquo outcomesby birth cohort quantile and gender in proportion to exposuretime We believe that plausible omitted variables are unlikely to



have all of these properties and therefore conclude that our base-line estimate of 4 convergence in outcomes per year of childhoodexposure to an area is unbiased

Our findings yield three broad lessons First place mattersfor intergenerational mobility the differences we see in outcomesacross neighborhoods are largely due to the causal effect of placesrather than differences in the characteristics of their residentsSecond place matters largely because of differences in child-hood environment rather than the differences in labor marketconditions that have received attention in previous studies ofplace Moving to a better area just before entering the labor mar-ket has little impact on individualrsquos outcomes suggesting thatplace-conscious policies to promote upward mobility should focusprimarily on improving the local childhood environment ratherthan conditions in adulthood Third each year of childhood expo-sure matters roughly equally there is no ldquocritical agerdquo after whichthe returns to living in a better neighborhood fall sharply This re-sult is germane to recent policy discussions regarding early child-hood interventions as it suggests that improvements in neighbor-hood environments can be beneficial even in adolescence

Our results help explain why previous experimental studiesmdashmost notably the Moving to Opportunity (MTO) experimentmdashfailed to detect significant effects of moving to a betterneighborhood on economic outcomes Prior analyses of the MTOexperiment focused primarily on the effects of neighborhoods onadults and older youth (eg Kling Liebman and Katz 2007) be-cause data on the long-term outcomes of younger children wereunavailable In a companion paper (Chetty Hendren and Katz2016) we link the MTO data to tax records and show that theMTO data exhibit childhood exposure effects consistent with thoseidentified here In particular Chetty Hendren and Katz (2016)find substantial improvements in earnings and other outcomes forchildren whose families received experimental vouchers to moveto low-poverty neighborhoods at young ages In contrast childrenwho moved at older ages experienced no gains or slight losses7

7 One important distinction between the two studies is that the analysissample in the present quasi-experimental study consists entirely of families whomoved across CZs whereas the MTO experiment compares families who movedto lower poverty neighborhoods with families who did not move at all or movedwithin an area similar to where they lived before As a result the analysis hereidentifies the effects of moving to better versus worse areas conditional on movingto a different area whereas the MTO analysis compares the effects of moving to alower-poverty neighborhood versus staying in a higher-poverty area The exposure



More generally our findings imply that much of theneighborhood-level variation in economic outcomes documentedin previous observational studies does in fact reflect causal effectsof place but that such effects arise through accumulated childhoodexposure rather than immediate impacts on adults The idea thatexposure time to better neighborhoods may matter has been notedsince at least Wilson (1987) and Jencks and Mayer (1990) and hasreceived growing attention in observational studies in sociology(Crowder and South 2011 Wodtke Harding and Elwert 2011Wodtke 2013 Sharkey and Faber 2014) We contribute to this lit-erature by presenting quasi-experimental estimates of exposureeffects addressing the concerns about selection and omitted vari-able bias that arise in observational studies (Ludwig et al 2008)Although we find evidence of childhood exposure effects that arequalitatively consistent with the observational studies we findno evidence of exposure effects in adulthood in this study or ourMTO study contrary to the patterns observed in observationaldata (Clampet-Lundquist and Massey 2008)

Our findings are also consistent with recent studies that useother research designsmdashrandom assignment of refugees (Dammand Dustmann 2014) housing demolitions (Chyn forthcoming)and selection corrections using group characteristics (Altonji andMansfield 2016)mdashto show that neighborhoods have causal effectson childrenrsquos long-term outcomes The present analysis comple-ments these studies and Chetty Hendren and Katzrsquos (2016) re-analysis of the MTO experiment in two ways First it sheds lighton the mechanisms underlying neighborhood effects by deliver-ing precise estimates of the magnitude and linear age pattern ofchildhood exposure effects Second it develops a scalable methodto estimate neighborhood effects in all areas even those whererandomized or natural experiments are unavailable8

effect estimates here net out any fixed disruption costs of moving to a differenttype of area whereas such costs are not netted out in the MTO experiment Thisdistinction may explain why Chetty Hendren and Katz (2016) find slightly neg-ative effects for children who move at older ages in the MTO data whereas weestimate positive exposure effects of moving to a better area (conditional on mov-ing) at all ages here

8 Our estimates of neighborhood exposure effects are based on householdswho choose to move to certain areas The effects of moving a randomly selectedhousehold to a new area may differ since households that choose to move to agiven area may be more likely to benefit from that move The fact that exposureeffects are similar within the subset of displaced households and are symmetric formoves to better and worse areas suggests such heterogeneity in exposure effects is



This article is organized as follows Section II describes thedata Section III presents our empirical framework starting witha description of differences in intergenerational mobility acrossareas and then specifying our estimating equations Section IVpresents baseline estimates of neighborhood exposure effects anddiscusses the mechanisms through which neighborhoods affectchildrenrsquos incomes Section V presents tests evaluating our iden-tification assumption Section VI presents estimates of exposureeffects for other outcomes Section VII concludes Supplementaryresults and details on estimation methodology are provided in theOnline Appendix

II DATA

We use data from federal income tax records spanning 1996ndash2012 The data include both income tax returns (1040 forms) andthird-party information returns (eg W-2 forms) which containinformation on the earnings of those who do not file tax returnsBecause our empirical analysis is designed to determine howmuch of the geographic variation in intergenerational mobilitydocumented by Chetty et al (2014) is due to causal effects ofplace we use an analysis sample that is essentially identical tothe ldquoextended samplerdquo used in Chetty et al (2014) Online Ap-pendix A of Chetty et al (2014) gives a detailed description ofhow we construct the analysis sample starting from the raw pop-ulation data Here we briefly summarize the key variable andsample definitions following Section III of Chetty et al (2014)

IIA Sample Definitions

Our base data set of children consists of all individuals who (i)have a valid Social Security Number or Individual Taxpayer Iden-tification Number (ii) were born between 1980ndash1988 and (iii) areUS citizens as of 20139 We impose the citizenship requirementto exclude individuals who are likely to have immigrated to theUnited States as adults for whom we cannot measure parent in-come We cannot directly restrict the sample to individuals born

limited but further work is needed to understand how exposure effects vary withhouseholdsrsquo willingness to move

9 For selected outcomes that can be measured at earlier ages such as teenagelabor force participation rates we extend the sample to include more recent birthcohorts up to 1996



in the United States because the database only records currentcitizenship status

We identify the parents of a child as the first tax filers (be-tween 1996ndash2012) who claim the child as a dependent and werebetween the ages of 15 and 40 when the child was born10 If thechild is first claimed by a single filer the child is defined as havinga single parent For simplicity we assign each child a parent (orparents) permanently using this algorithm regardless of any sub-sequent changes in parentsrsquo marital status or dependent claiming

If parents never file a tax return we do not link them totheir child Although some low-income individuals do not file taxreturns in a given year almost all parents file a tax return atsome point between 1996 and 2012 to obtain a tax refund on theirwithheld taxes and the Earned Income Tax Credit (Cilke 1998)As a result approximately 94 of the children in the 1980ndash1988birth cohorts are claimed as a dependent at some point between1996 and 2012 The fraction of children linked to parents dropssharply prior to the 1980 birth cohort because our data begin in1996 and many children begin to the leave the household startingat age 17 (Chetty et al 2014 Online Appendix Table I) This iswhy we limit our analysis to children born during or after 1980

Our full analysis sample includes all children in the base dataset who are born in the 1980ndash1988 birth cohorts for whom we areable to identify parents and whose mean parent income between1996 and 2000 is strictly positive11 We divide the full sample intotwo parts permanent residents (or stayers) and movers We definethe permanent residents of each CZ c as the subset of parents whoreside in a single CZ c in all years of our sample 1996ndash2012 Themovers sample consists of individuals in the full sample who arenot permanent residents

In our baseline analysis we focus on the subset of individualswho live in CZs with populations in the 2000 census above 250000(excluding 196 of the observations) to ensure that we have ade-quately large samples to estimate permanent residentsrsquo outcomes(the key independent variables in our analysis) precisely There

10 We impose the 15ndash40 age restriction to limit links to grandparents or otherguardians who might claim a child as a dependent

11 We limit the sample to parents with positive income (excluding 15 ofchildren) because parents who file a tax return as is required to link them to achild yet have zero income are unlikely to be representative of individuals withzero income and those with negative income typically have large capital losseswhich are a proxy for having significant wealth



are approximately 246 million children in the baseline analysissample for whom we observe outcomes at age 24 or later of whom195 million are children of permanent residents

IIB Variable Definitions and Summary Statistics

In this section we define the key variables we use in our anal-ysis We measure all monetary variables in 2012 dollars adjustingfor inflation using the headline consumer price index (CPI-U) Webegin by defining the two key variables we measure for parentsincome and location

1 Parent Income Our primary measure of parent income istotal pretax income at the household level which we label par-ent family (or household) income In years where a parent filesa tax return we define family income as adjusted gross income(as reported on the 1040 tax return) plus tax-exempt interestincome and the nontaxable portion of Social Security and Disabil-ity (SSDI) benefits In years where a parent does not file a taxreturn we define family income as the sum of wage earnings (re-ported on form W-2) unemployment benefits (reported on form1099-G) and gross SSDI benefits (reported on form SSA-1099) forboth parents12 In years where parents have no tax return and noinformation returns family income is coded as 013 Income is mea-sured prior to the deduction of income taxes and employee-levelpayroll taxes and excludes nontaxable cash transfers and in-kindbenefits

In our baseline analysis we average parentsrsquo family incomeover the five years from 1996 to 2000 to obtain a proxy for par-ent lifetime income that is less affected by transitory fluctuations(Solon 1992) We use the earliest years in our sample to best re-flect the economic resources of parents while the children in oursample are growing up14 Because we measure parent income in a

12 The database does not record W-2s and other information returns prior to1999 so a nonfilerrsquos income is coded as 0 prior to 1999 Assigning nonfiling parents0 income has little impact on our estimates because only 31 of parents in thefull analysis sample do not file in each year prior to 1999 and most nonfilers havevery low W-2 income (Chetty et al 2014) For instance in 2000 the median W-2income among nonfilers in our baseline analysis sample was $0

13 Importantly these observations are true zeros rather than missing dataBecause the database covers all tax records we know that these individuals have0 taxable income

14 Formally we define mean family income as the motherrsquos family incomeplus the fatherrsquos family income in each year from 1996 to 2000 divided by 10 (ordivided by 5 if we only identify a single parent) For parents who do not change



fixed set of years the age of the child when parent income is mea-sured varies across birth cohorts We account for this variation byconditioning on the childrsquos birth cohort throughout our analysis

2 Parent Location In each year parents are assigned ZIPcodes of residence based on the ZIP code from which they filedtheir tax return If the parent does not file in a given year wesearch W-2 forms for a payee ZIP code in that year Nonfilerswith no information returns are assigned missing ZIP codes Forchildren whose parents were married when they were first claimedas dependents we always track the motherrsquos location if maritalstatus changes We map parentsrsquo ZIP codes to counties and CZsusing the crosswalks and methods described in Chetty et al (2014Online Appendix A)

Next we define the outcomes that we analyze for children

3 Income We define child family income in the same way asparent family income We measure childrenrsquos annual incomes atages ranging from 24ndash30 and define the childrsquos household basedon his or her marital status at the point at which income is mea-sured For some robustness checks we analyze individual incomedefined as the sum of individual W-2 wage earnings unemploy-ment insurance benefits SSDI payments and half of householdself-employment income

4 Employment We define an indicator for whether the childis employed at a given age based on whether he or she has aW-2 form filed on his or her behalf at that age We measure em-ployment rates starting at age 16 to analyze teenage labor forceparticipation

5 College Attendance We define college attendance as anindicator for having one or more 1098-T forms filed on onersquosbehalf when the individual is aged 18ndash23 Title IV institutionsmdashall colleges and universities as well as vocational schools andother postsecondary institutions eligible for federal student

marital status this is simply mean family income over the five-year period Forparents who are married initially and then divorce this measure tracks the meanfamily incomes of the two divorced parents over time For parents who are sin-gle initially and then get married this measure tracks individual income prior tomarriage and total family income (including the new spousersquos income) after mar-riage These household measures of income increase with marriage and naturallydo not account for cohabitation to ensure that these features do not generate biaswe assess the robustness of our results to using individual measures of income



aidmdashare required to file 1098-T forms that report tuition pay-ments or scholarships received for every student The 1098-Tforms are available from 1999 to 2012 and are filed directly bycolleges independent of whether an individual files a tax returnComparisons to other data sources indicate that 1098-T formscapture more than 95 of college enrollment in the United States(Chetty et al 2017)

6 Teenage Birth For both males and females we identifya birth as being listed as a parent on a childrsquos application for asocial security number using data from the Social Security Ad-ministrationrsquos Kidlink (DM-2) database We define a teenage birthas having a child between the ages of 13 and 1915

7 Marriage We define an indicator for whether the child ismarried at a given age based on the marital status listed on 1040forms for tax filers We code nonfilers as single because linkedCPSIRS data show that the vast majority of nonfilers below theage of 62 are single (Cilke 1998)

8 Summary Statistics Table I reports summary statistics forour analysis sample and various subgroups used in our CZ-levelanalysis Online Appendix Table I presents analogous statisticsfor the sample used in our county-level analysis The first panelreports statistics for permanent residents in our full analysis sam-ple who live in CZs with more than 250000 people The secondpanel considers the 44 million children who moved between CZswith more than 250000 people (excluding children whose parentsmoved more than three times between 1996 and 2012 who ac-count for 3 of the observations) The third panel focuses on ourprimary analysis sample of one-time movers children whose par-ents moved exactly once across CZs between 1996 and 2012 areobserved in the destination CZ for at least two years and moved at

15 The total count of births in the SSA DM-2 database closely matches vitalstatistics counts from the Centers for Disease Control and Prevention prior to 2008however the DM-2 database contains approximately 10 fewer births between2008 and 2012 Using an alternative measure of teenage birth that does not sufferfrom this missing data problemmdashin which we define a person as having a teenbirth if he or she ever claims a dependent who was born while he or she wasbetween the ages of 13 and 19mdashyields very similar results (not reported) We donot use the dependent-claiming definition as our primary measure of teenage birthbecause it only covers children who are claimed as dependents



TABLE ISUMMARY STATISTICS FOR CZ PERMANENT RESIDENTS AND MOVERS

Mean Std dev Median Num of obsVariable (1) (2) (3) (4)

Panel A Permanent residents Families who do not move across CZsParent family income 89909 357194 61300 19499662Child family income at 24 24731 140200 19600 19499662Child family income at 26 33723 161423 26100 14894662Child family income at 30 48912 138512 35600 6081738Child individual income at 24 20331 139697 17200 19499662Child married at 26 025 043 000 12997702Child married at 30 039 049 000 6081738Child attends college between 18ndash23 070 046 100 17602702Child has teen birth (females only) 011 032 000 9670225Child working at age 16 041 049 000 13417924

Panel B Families who move 1ndash3 times across CZsParent family income 90468 376413 53500 4374418Child family income at 24 23489 57852 18100 4374418Child family income at 26 31658 99394 23800 3276406Child family income at 30 46368 107380 32500 1305997Child individual income at 24 19091 51689 15600 4374418Child married at 26 025 043 000 2867598Child married at 30 038 049 000 1305997Child attends college between 18ndash23 066 047 100 3965610Child has teen birth (females only) 013 033 000 2169207Child working at age 16 040 049 000 3068421

Panel C Primary analysis sample families who move exactly once across CZsParent family income 97064 369971 58700 1553021Child family income at 24 23867 56564 18600 1553021Child family income at 26 32419 108431 24500 1160278Child family income at 30 47882 117450 33600 460457Child individual income at 24 19462 48452 16000 1553021Child married at 26 025 043 000 1016264Child married at 30 038 049 000 460457Child attends college between 18ndash23 069 046 100 1409007Child has teen birth (females only) 011 032 000 769717Child working at age 16 039 049 000 1092564

Notes The table presents summary statistics for the samples used in our CZ-level analyses The fullanalysis sample of children consists of all individuals in the tax data who (i) have a valid Social SecurityNumber or Individual Taxpayer Identification Number (ii) were born between 1980 and 1988 and (iii) areUS citizens as of 2013 We report summary statistics for three subsets of this sample Panel A shows statisticsfor permanent residentsmdashchildren whose parents do not move across CZs throughout our sample window(1996 to 2012)mdashwho live in CZs with more than 250000 people based on the 2000 census Panel B showsstatistics for families who moved once twice or three times across CZs with more than 250000 people from1996ndash2012 Panel C shows statistics for our primary analysis sample children whose families moved exactlyonce across CZs with more than 250000 people are observed in the destination CZ for at least two years andmoved at least 100 miles (based on their ZIP codes) Parent family income is the average pretax householdincome from 1996 to 2000 measured as AGI plus tax-exempt interest income and the nontaxable portion ofSocial Security and Disability (SSDI) benefits for tax-filers and using information returns for nonfilers Childfamily income is measured analogously at various ages while child individual income is defined as the sumof individual W-2 wage earnings unemployment insurance benefits SSDI payments and half of householdself-employment income Marital status is defined based on the marital status listed on 1040 forms for taxfilers nonfilers are coded as single College attendance is defined as having a 1098-T form filed on onersquosbehalf at any point between the ages of 18 and 23 Teenage birth is defined (for women only) as having achild between the ages of 13 and 19 using data from the Social Security Administrationrsquos DM-2 databaseWe define an indicator for working at age 16 based on having a W-2 form filed on onersquos behalf at that ageAll dollar values are reported in 2012 dollars deflated using the CPI-U See Section II for further details onvariable and sample definitions



least 100 miles16 There are 16 million children in this one-timemovers sample

Although our analysis does not require movers to be compa-rable to permanent residents we find that movers and perma-nent residents have similar characteristics Median parent familyincome is $61300 for permanent residents compared to $58700for one-time movers Children of permanent residents have a me-dian family income of $35600 when they are 30 years old com-pared with $33600 for one-time movers Roughly 70 of childrenof permanent residents and one-time movers are enrolled in a col-lege at some point between the ages of 18 and 23 Eleven percentof daughters of permanent residents and one-time movers have ateenage birth

III EMPIRICAL FRAMEWORK

In this section we first present a descriptive characteriza-tion of intergenerational mobility for children who grow up indifferent areas in the United States We then formally define ourestimands of interestmdashchildhood exposure effectsmdashand describethe research design we use to identify these exposure effects inobservational data

IIIA Geographical Variation in Outcomes of PermanentResidents

We conceptualize ldquoneighborhoodrdquo effects as the sum of placeeffects at different geographies ranging from broad to narrowCZ counties ZIP codes and census tracts In the main text ofthis article we focus on variation across CZs CZs are aggrega-tions of counties based on commuting patterns in the 1990 censusconstructed by Tolbert and Sizer (1996) There are 741 CZs in theUnited States on average each CZ contains four counties andhas a population of 380000 We replicate the results reported inthe main text at the county level in Online Appendix C We focuson variation across relatively broad geographic units to maximizestatistical precision because some of our research designs require

16 We impose these restrictions to eliminate moves across CZ borders thatdo not reflect a true change of location We measure the distance of moves asthe distance between the centroids of the origin and destination ZIPs (obtainedfrom wwwboutellcomzipcodes) We show the robustness of our results to usingalternative cutoffs for minimum population size and move distances in OnlineAppendix A



large sample sizes to discern fine variation in permanent resi-dentsrsquo outcomes across subsamples

We characterize the outcomes of children who spent theirentire childhoods in a single CZ by focusing on children of ldquoper-manent residentsrdquo mdash parents who stay in the same CZ between1996 and 201217 Importantly our definition of permanent resi-dents conditions on parentsrsquo locations not childrenrsquos locations inadulthood The CZ where a child grew up may differ from the CZwhere he lives when we measure his earnings in adulthood

Since places can have different effects across parent incomelevels and over time we characterize childrenrsquos mean outcomesconditional on their parentsrsquo income separately for each CZ c andbirth cohort s Chetty et al (2014) show that measuring incomesusing percentile ranks (rather than dollar levels) has significantstatistical advantages Following their approach we define childirsquos percentile rank yi based on his position in the national distribu-tion of incomes relative to all others in his birth cohort Similarlywe measure the percentile rank of the parents of child i p(i) basedon their positions in the national distribution of parental incomefor child irsquos birth cohort

Let ypcs denote the mean rank of children with parents at per-centile p of the income distribution in CZ c in birth cohort s Fig-ure I illustrates how we estimate ypcs for children born in 1980 toparents who are permanent residents of the Chicago CZ This fig-ure plots the mean child rank at age 30 within each percentile binof the parent income distribution E[yi|p(i) = p] The conditionalexpectation of a childrsquos rank given his parentsrsquo rank is almost per-fectly linear a property that is robust across CZs (Chetty et al2014 Online Appendix Figure IV) Exploiting linearity we parsi-moniously summarize the relationship between childrenrsquos mean

17 Because our data start in 1996 we cannot measure parentsrsquo location overtheir childrenrsquos entire childhood For the 1980 birth cohort we measure parentsrsquolocation between the ages of 16 and 32 for the 1991 birth cohort we measureparentsrsquo location between 5 and 21 This creates measurement error in childrenrsquoschildhood environment that is larger in earlier birth cohorts Fortunately we findthat our results do not vary significantly across birth cohorts and in particular re-main similar for the most recent birth cohorts The reason such measurementerror turns out to be modest empirically is that most families who stay in agiven area for several years tend not to have moved in the past either For ex-ample among families who stayed in the same CZ c when their children werebetween ages 16 and 24 815 of them lived in the same CZ when their childrenwere age 8



FIGURE I

Mean Child Income Rank versus Parent Income Rank for Children Raised inChicago

This figure presents a binned scatter plot of the relationship between childrenrsquosincome ranks and parent income ranks for children raised in Chicago The pointson the figure plot the mean rank of children within each parental income percentilebin The best-fit line is estimated using an OLS regression on the underlying microdata The figure also reports the slope of the best-fit line (the rank-rank slope)along with the standard error of the estimate (in parentheses) The sample includesall children in the 1980 birth cohort in our analysis sample whose parents werepermanent residents of the Chicago commuting zone during the sample period(1996ndash2012) Childrenrsquos incomes are measured at the household (ie family) levelat age 30 parentsrsquo incomes are defined as mean family income from 1996 to 2000Children are assigned ranks based on their incomes relative to all other childrenin their birth cohort Parentsrsquo are assigned ranks based on their incomes relativeto other parents of children in the same birth cohort

income ranks and their parentsrsquo ranks by regressing childrenrsquosranks on their parentsrsquo ranks in each CZ c and birth cohort s

(1) yi = αcs + ψcs pi + εi

We then estimate ypcs using the fitted values from this regression

(2) ypcs = αcs + ψcs p

For example in Chicago y25c1980 = 401 for children growing upat the 25th percentile of the national income distribution andy75c1980 = 593 for children growing up at the 75th percentile

Figure II maps childrenrsquos mean income ranks at age 30 byCZ for children with parents at the 25th percentile (Panel A) and



FIGURE II

Mean Income Ranks for Children of Permanent Residents

These maps plot childrenrsquos mean percentile ranks at age 30 conditional on havingparents at the 25th percentile (Panel A) and 75th percentile (Panel B) The mapsare constructed by grouping CZs into 10 deciles and shading the areas so thatlighter colors correspond to higher outcomes for children Areas with fewer than10 children for which we have insufficient data to estimate outcomes are shadedwith the striped pattern The sample includes all children in the 1980 birth cohortin our analysis sample whose parents are permanent residents (ie whose parentsdo not move across CZs between 1996 and 2012) To construct these estimates wefirst regress childrenrsquos family income ranks on a constant and their parentsrsquo familyincome ranks separately for each CZ and birth cohort We then define the predictedincome rank for children with parents at percentile p in CZ c in birth cohort s (ypcs)as the intercept + p times the slope of this regression Panel A reports the predictedchild rank for parents at p = 25 which corresponds to an annual household incomeof $30000 Similarly Panel B reports the predicted child rank for parents at p = 75which corresponds to an annual household income of $97000 See notes to Figure Ifor details on definitions of parent and child income ranks



75th percentile (Panel B) analogous maps at the county levelare presented in Online Appendix Figure I We construct thesemaps by dividing CZs into deciles based on their estimated valueof y25cs and y75cs with lighter colors representing deciles withhigher mean incomes As documented by Chetty et al (2014)childrenrsquos incomes vary substantially across CZs especially forchildren from low-income families Chetty et al (2014 SectionVC) discuss the spatial patterns in these maps in detail Herewe focus on investigating whether the variation in these maps isdriven by causal effects of place or heterogeneity in the types ofpeople living in different places

IIIB Definition of Exposure Effects

Our objective is to determine how much a childrsquos potential out-comes would improve on average if he were to grow up in an areawhere the permanent residentsrsquo outcomes are 1 percentile pointhigher We answer this question by studying children who moveacross areas to estimate childhood exposure effects We define theexposure effect at age m as the impact of spending year m of onersquoschildhood in an area where permanent residentsrsquo outcomes are 1percentile point higher

Formally consider a hypothetical experiment in which werandomly assign children to new neighborhoods d starting at agem for the rest of their childhood The best linear predictor of chil-drenrsquos outcomes yi in the experimental sample based on the per-manent residentsrsquo outcomes in CZ d (ypds) can be written as

(3) yi = αm + βmypds + θi

where the error term θ i captures family inputs and other deter-minants of childrenrsquos outcomes Since random assignment guar-antees that θ i is orthogonal to ypds estimating equation (3) usingOLS yields a coefficient βm that represents the mean impact ofspending year m of onersquos childhood onward in an area where per-manent residents have 1 percentile better outcomes We definethe exposure effect at age m as γ m = βm minus βm+118 Note that ifincome yi is measured at age T βm = 0 for m gt T as moving

18 For simplicity we do not allow βm to vary across parent income percentilesp in our baseline analysis thereby estimating the average exposure effect acrossfamilies with different incomes We estimate equation (3) separately by parentalincome level in Online Appendix Table III



after the outcome is measured cannot have a causal effect on theoutcome

Estimating the exposure effects γ m is of interest for severalreasons First a positive effect (at any age) allows us to reject thenull hypothesis that neighborhoods do not matter a null of inter-est given prior experimental evidence Second γ m is informativeabout the ages at which neighborhood environments matter mostfor childrenrsquos outcomes Third the magnitude of β0 = sumT

t=0 γm (theimpact of assigning children to a better neighborhood from birth)provides an estimate of the degree to which the differences inchildrenrsquos outcomes across areas are due to place effects versusselection If place effects are homogeneous across children withinbirth cohorts and parent income groups β0 = 0 would imply thatall of the variation across areas is due to selection and β0 = 1would imply that all of the variation reflects causal effects ofplace More generally the magnitude of β0 tells us how muchof the differences across areas in Figure II rub off on children whoare randomly assigned to live there from birth

Although β0 sheds light on the causal effect of places on av-erage we caution that it does not identify the causal effect ofany given area on a childrsquos potential outcomes The causal ef-fect of growing up in a given CZ c will generally differ from themean predicted impact based on permanent residentsrsquo outcomes(α + β0 ypds) because selection and causal effects will vary acrossareas We build on the methodology developed in this article toestimate the causal effect of each CZ and county in the secondarticle in this series (Chetty and Hendren 2018a)

IIIC Estimating Exposure Effects in Observational Data

We estimate exposure effects by studying families who moveacross CZs with children of different ages in observational data Inobservational data the error term θ i in equation (3) will generallybe correlated with ypds For instance parents who move to a goodarea may have latent ability or wealth that produces better childoutcomes Estimating equation (3) in an observational sample offamilies who move exactly once yields a regression coefficient

bm = βm + δm

where δm = cov(θi ypds)var(ypds)

is a standard selection effect that measuresthe extent to which parental inputs and other determinants ofchildrenrsquos outcomes for movers covary with permanent residentsrsquo



outcomes Fortunately the identification of exposure effects doesnot require that where people move is orthogonal to a childrsquos po-tential outcomes Instead it requires that when people move tobetter versus worse areas is orthogonal to a childrsquos potential out-comes as formalized in the following assumption

ASSUMPTION 1 Selection effects do not vary with the childrsquosage at move δm = δ for all m

Assumption 1 allows for the possibility that the families whomove to better areas may differ from those who move to worseareas but requires that the extent of such selection does not varywith the age of the child when the parent moves Under thisassumption we obtain consistent estimates of exposure effectsγ m = βm minus βm+1 = bm minus bm+1 from equation (3) even in observa-tional data because the selection effect δ cancels out when estimat-ing the exposure effect We can estimate the selection effect δ itselfby examining the outcomes of children whose families move aftertheir income is measured for example at age a 30 if incomeis measured at age T = 30 Because moves at age a gt T cannothave a causal effect on childrenrsquos outcomes at age 30 bm = δ form gt T under Assumption 1 Using the estimated selection effectwe can identify the causal effect of moving to a better area atage m as βm = bm minus bT+1 and thereby identify β0 the causal ef-fect of growing up from birth in an area with 1 percentile betteroutcomes

Of course Assumption 1 is a strong restriction that may nothold in practice We therefore evaluate its validity in detail afterpresenting a set of baseline estimates in the next section

IV BASELINE ESTIMATES OF CHILDHOOD EXPOSURE EFFECTS

This section presents our baseline estimates of exposure ef-fects γ m We begin by presenting a set of semiparametric esti-mates of γ m using specifications that condition on origin fixedeffects and correspond most closely to the hypothetical experi-ment described in Section IIIB We then present estimates fromparametric models that show how moversrsquo outcomes can be parsi-moniously modeled as a linear combination of the outcomes of per-manent residents in origins and destination Finally we presenta set of supplementary results that shed light on the mechanismsthrough which neighborhoods affect childrenrsquos outcomes

In our baseline analysis we focus on children whose parentsmoved across CZs exactly once between 1996 and 2012 and are



observed in the destination CZ for at least two years We alsorestrict our attention to families who moved at least 100 miles toexclude moves across CZ borders that do not reflect a true changeof neighborhood and limit the sample to CZs with populationsabove 250000 to minimize sampling error in the estimates ofpermanent residentsrsquo outcomes ypds We show that the findings arerobust to alternative cutoffs for population size and move distancein Online Appendix A and present estimates that include familieswho move more than once in Online Appendix B

In prior work (Chetty et al 2014) we found that the inter-generational correlation between parentsrsquo and childrenrsquos incomesstabilizes when children turn 30 as college graduates experiencesteeper wage growth in their twenties (Haider and Solon 2006)Measuring income at age 30 limits us to estimating exposure ef-fects only after age 15 given the time span of our data set19 For-tunately measuring income at earlier ages (from 24 to 30) turnsout not to affect the exposure effect estimates The reason is thatour estimates of bm are identified by correlating the incomes ofchildren who move with the incomes of permanent residents inthe destination at the same age in adulthood This property ofour estimator allows us to measure the degree of convergence ofmoversrsquo outcomes to permanent residentsrsquo outcomes at any ageirrespective of whether childrenrsquos incomes at that age reflect theirpermanent incomes For example if a given area c sends manychildren to college and therefore generates relatively low incomesat age 24 we would obtain a higher estimate of bm if a child whomoves to area c has a lower level of income at age 24 We there-fore measure income at age 24 in our baseline specifications toestimate exposure effects for the broadest range of ages20

IVA Semiparametric Estimates

To begin consider the set of children whose familiesmoved when they were exactly m years old We analyze howthese childrenrsquos incomes in adulthood are related to those of

19 The most recent birth cohort for which we observe income at age 30 (in2012) is the 1982 cohort because our data begin in 1996 we cannot observe movesbefore age 15

20 We show below that we obtain similar estimates when measuring incomeat later ages (from 26 to 30) over the overlapping range of ages at which childrenmove We do not study income before age 24 because we find that exposure effectspersist until age 23 when income is measured at any point between 24 and 30



permanent residents in their destination CZ using the followinglinear regression

(4) yi = αqos + bmodps + ε1i

where yi denotes the childrsquos income rank at age 24 αqos is a fixedeffect for the origin CZ o by parent income decile q by birth co-hort s and odps = ypds minus ypos is the difference in predicted incomerank (at age 24) of permanent residents in the destination ver-sus origin for the relevant parent income rank p and birth cohorts Equation (4) can be interpreted as an observational analog ofthe specification in equation (3) that we would ideally estimate inexperimental data21

Figure III presents a nonparametric binned scatter plot cor-responding to the regression in equation (4) for children who moveat age m = 13 To construct the figure we first demean both yi andodps within the parent decile (q) by origin (o) by birth cohort (s)cells in the sample of movers at age m = 13 to construct residualsyr

i = yi minus E[yi|q o s] and rodps = odps minus E[odps|q o s] We then

divide the rodps residuals into 20 equal-size groups (ventiles) and

plot the mean value of yri versus the mean value of r

odps in eachbin

Figure III shows that children who move to areas where chil-dren of permanent residents earn more at age 24 themselves earnmore when they are 24 The relationship between yi and odps islinear The regression coefficient of b13 = 0615 estimated in themicrodata using equation (4) implies that a 1 percentile increasein ypds is associated with a 0615 percentile increase in yi for thechildren who move at age 13

Building on this approach we estimate analogous regressioncoefficients bm for children whose parents move at each age mfrom 9 to 30 We estimate bm using the following specification(5)

yi = αqosm +30sum

m=9

bmI(mi = m)odps +1987sum

s=1980

κs I(si = s)odps + ε2i

21 We use parent income deciles rather than percentiles to define the fixedeffects αqos to simplify computation using finer bins to measure parent incomegroups has little effect on the estimates Conditional on parent percentile originand birth cohort the variation in odps is entirely driven by variation in thedestination outcomes (ypds) Hence bm is identified from variation in ypds as inequation (3) up to the approximation error from using parent deciles instead ofexact percentiles



FIGURE III

Moversrsquo Outcomes versus Predicted Outcomes Based on Permanent Residents inDestination

This figure presents a binned scatter plot depicting the relationship between theincome ranks of children who moved to a different CZ at age 13 and the differencesin the outcomes of permanent residents in the destination versus origin CZ Thesample includes all children in the 1980ndash1988 birth cohorts whose parents movedwhen the child was 13 years old and moved only once between 1996 and 2012Childrenrsquos family income ranks yi are measured at age 24 Permanent residentsrsquopredicted ranks for each parent income percentile p CZ c and birth cohort s(ypcs) are constructed using the methodology described in the notes to Figure I Toconstruct the figure we demean both yi and odps = ypds minus ypos within the parentdecile (q) by origin (o) by birth cohort (s) cells in the sample of movers at age m = 13to construct residuals yr

i = yi minus E[yi |q o s] and rodps = odps minus E[odps|q o s]

We then divide the rodps residuals into 20 equal-size groups (ventiles) and plot

the mean value of yri versus the mean value of r

odps in each bin The slope of thebest-fit line which corresponds to b13 in equation (4) is estimated using an OLSregression on the underlying microdata with standard errors in parentheses

where αqosm is an origin CZ by parent income decile by birth cohortby age at move fixed effect and I(xi = x) is an indicator functionthat is 1 when xi = x and 0 otherwise This specification gener-alizes equation (4) by fully interacting the age at move m withthe independent variables in equation (4) In addition we permitthe effects of odps to vary across birth cohorts (captured by theκs coefficients) because our ability to measure parentsrsquo locationsduring childhood varies across birth cohorts We observe childrenrsquoslocations starting only at age 16 for the 1980 cohort but startingat age 8 for the 1988 cohort This leads to greater measurementerror in odps for earlier birth cohorts which can confound our



estimates of bm because the distribution of ages at move is unbal-anced across cohorts (see Online Appendix A for further details)By including cohort interactions we identify bm from within-cohort variation in ages at move22

Figure IV Panel A plots estimates of bm from equation (5)The estimates exhibit two key patterns selection effects after age24 and exposure effects before age 24 First the fact that bm gt 0for m gt 24 is direct evidence of selection effects (δm gt 0) as movesafter age 24 cannot have a causal effect on income at 24 Familieswho move to better areas have children with better unobservableattributes The degree of selection δm does not vary significantlywith m above age 24 regressing bm on m for m 24 yields astatistically insignificant slope of 0001 (std err = 0011) Thisresult is consistent with Assumption 1 which requires that selec-tion does not vary with the childrsquos age at move The mean valueof δm for m 24 is δ = 0126 that is families who move to anarea where permanent residents have 1 percentile better out-comes have 0126 percentile better outcomes themselves purelydue to selection effects Assumption 1 allows us to extrapolate theselection effect of δ = 0126 back to earlier ages m lt 24 as shownby the dashed horizontal line in Figure IV Panel A and therebyidentify causal exposure effects at earlier ages

This leads to the second key pattern in Figure IV Panel Awhich is that the estimates of bm decline steadily with the age atmove m for m lt 24 Under Assumption 1 this declining patternconstitutes evidence of an exposure effect that is that movingto a better area earlier in childhood generates larger long-termgains23 The linearity of the relationship between bm and the ageat move m in Figure IV Panel A below age 24 implies that theexposure effect γ m = bm+1 minus bm is approximately constant withrespect to age at move m Regressing bm on m for m lt 24 weestimate an average annual exposure effect of γ = 0044 (std err= 0003) That is the outcomes of children who move converge to

22 To avoid collinearity we omit the most recent cohort interaction with odps(the 1988 cohort when income is measured at age 24) We show below that thesecohort interactions have little impact on the estimates obtained from equation (5)but play a larger role in specifications that include family fixed effects

23 This declining pattern could also potentially be generated by critical ageeffects rather than effects that operate in proportion to exposure time We presentevidence in Section IVC supporting the interpretation of these results as exposureeffects



FIGURE IV

Childhood Exposure Effects on Income Ranks in Adulthood

Panel A plots estimates of the coefficients bm versus the childrsquos age when theparents move (m) using the semiparametric specification in equation (5) measur-ing childrenrsquos incomes at age 24 The sample includes all children in the primaryanalysis sample whose parents moved exactly once between 1996 and 2012 Thebm coefficients can be interpreted as the effect of moving to an area where per-manent resident outcomes are 1 percentile higher at age m They are estimatedby regressing the childrsquos income rank in adulthood yi on odps = ypds minus ypos thedifference between permanent residentsrsquo predicted ranks in the destination ver-sus the origin interacted with each age of the child at the time of the move mWe include origin CZ by parent income decile by birth cohort by age at movefixed effects when estimating this specification Panel B plots estimates from theparametric specification in equation (6) measuring childrenrsquos incomes at age 24This specification replicates the specification used in Panel A replacing the fixedeffects with indicators for the childrsquos age at the time of the move interacted withparent income rank and predicted outcomes for permanent residents in the origininteracted with birth cohort fixed effects The dashed vertical lines separate thedata into two groups age at move m 23 and m gt 23 Best-fit lines are estimatedusing unweighted OLS regressions of the bm coefficients on m separately for m 23 and m gt 23 The slopes of these regression lines are reported along withstandard errors (in parentheses) on the left side of each panel for m 23 andon the right side for m gt 23 The magnitudes of the slopes for m 23 representestimates of annual childhood exposure effects The parameter δ is defined as themean value of the bm estimates for m gt 23 this parameter represents a selectioneffect because moves after age 24 cannot affect income measured at age 24 InPanel A the dashed horizontal line shows the value of the selection effect δ theidentification assumption underlying the analysis is that the selection effect δ doesnot vary with the childrsquos age at move m

the outcomes of permanent residents of the destination area at arate of 44 per year of exposure until age 2324

24 Figure IV Panel A is identified from variation in moversrsquo destinationsholding their origin fixed An alternative approach is to exploit variation in originsholding destinations fixed Online Appendix Figure II presents estimates of bmidentified from variation in origins by replacing the origin (αqosm) fixed effects inequation (5) with destination (αqdsm) fixed effects The resulting estimates yield



Because some children do not move with their parents theestimates of bm in equation (5) should be interpreted as intent-to-treat (ITT) estimates in the sense that they capture the causal ef-fect of moving (plus the selection effect) for children whose parentsmoved at age m We can obtain treatment-on-the-treated (TOT)estimates for children who move with their parents by inflatingthe ITT estimates by the fraction of children who moved with theirparents at each age m25 In Online Appendix Figure III we showthat the TOT estimate of the exposure effect is γ TOT = 0040 Thisestimate is very similar to our baseline estimate because virtu-ally all children move with their parents below age 18 and roughly60 of children move with their parents between ages 18 and 23Because the treatment effects converge toward zero as the age atmove approaches 23 inflating the coefficients by 1

06 at later ageshas little impact on exposure effect estimates

IVB Parametric Estimates

Equation (5) includes more than 200000 fixed effects (αqosm)making it difficult to estimate in smaller samples and introduceadditional controls such as family fixed effects As a more tractablealternative we estimate a model in which we control parametri-cally for the two key factors captured by the αqosm fixed effects (i)the quality of the origin location which we model by interactingthe predicted outcomes for permanent residents in the origin atparent income percentile pi with birth cohort fixed effects and (ii)disruption costs of moving that may vary with the age at moveand parent income which we model using age at move fixed ef-fects linearly interacted with parent income percentile pi This

a qualitative pattern that is the mirror image of those in Figure IV Panel A thelater the family moves to the destination the more the childrsquos outcomes match thepermanent residents in the origin up to age 23 The estimated exposure effect of0030 is smaller than the estimates above because we measure childrenrsquos originswith greater error than destinations as our location data is left-censored This iswhy we focus on variation in destinations in most of our specifications

25 We identify children who move with their parents based on whether theyever file a tax return receive a W-2 form or attend a college in the destination CZ



leads to the following regression specification

yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+30sum

m=9


s=1980

κds I(si = s)odps + ε3i(6)

The first two terms of this specification control for origin qualityand disruption effects The third term represents the exposureeffects of interest and the fourth consists of cohort interactionswith odps to control for differential measurement error acrosscohorts as in equation (5)26

Figure IV Panel B plots the coefficients bm obtained fromestimating equation (6) The coefficients are very similar to thoseobtained from the more flexible specification used to constructFigure IV Panel A Regressing the bm coefficients on m for m 23we obtain an average annual exposure effect estimate of γ = 0038(std err = 0002) This estimate is similar to that obtained fromthe fixed effects specification because controlling for the quality ofthe origin using the permanent residentsrsquo outcomes is adequate toaccount for differences in origin quality Put differently moversrsquooutcomes can be modeled as a weighted average of the outcomes ofpermanents residents in the origin and destination with weightsreflecting the amount of childhood spent in the two places

When measuring income at age 24 we cannot determinewhether bm stabilizes after age 24 because moving after age 24has no causal effect on income or because we measure income atthat point In Online Appendix Figure IV we replicate the anal-ysis measuring income at ages 26 28 and 30 in addition to age24 All of these series display very similar patterns of exposureeffects in the overlapping age ranges showing that our estimatesof bm are insensitive to the age at which we measure childrenrsquosincomes in adulthood In particular all four series decline linearly

26 In addition to having many fewer fixed effects this specification usesvariation in both the quality of the origin (ypos) and the destination (ypds) to identifybm In contrast the semiparametric model in equation (5) is identified purelyfrom variation in destinations because it includes origin fixed effects Estimating aparametric model that identifies bm from variation in destinations by controllingfor outcomes of permanent residents in the origin interacted with the age of thechild at the time of the move (

sum30m=9 bmI(mi = m)ypos) yields very similar estimates



at a rate of approximately γ = 004 until age 23 and are flat there-after These results imply that neighborhood exposure before age23 is what matters for income in subsequent years

The kink at age 23 motivates the baseline regression specifi-cation that we use for the rest of our analysis We parameterizeboth the exposure and selection effects shown in Figure IV lin-early replacing the nonparametric

sum30m=9 bmI (mi = m) odps term

in equation (6) with two separate lines above and below age 23

yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+1987sum

s=1980

κds I(si = s)odps + I(mi 23)

(b0 + (23 minus mi)γ

)odps

+ I(mi gt 23)(δ + (23 minus mi)δprime)odps + ε3i(7)

Estimating this specification directly in the microdata yields anaverage annual exposure effect γ = 0040 (std err = 0002) asshown in column (1) of Table II27

The estimates of γ are robust to alternative specificationsand sample definitions Columns (2) and (3) of Table II show thatestimating γ using data only up to age 18 or 23mdashthat is ex-cluding the data at older ages that identifies the selection effectin equation (7)mdashyields similar estimates of γ Column (4) showsthat excluding the cohort interactions

sum1988s=1980 I(si = s)α2

s ypos andsum1987s=1980 κd

s I(si = s)odps in equation (7) does not affect the esti-mate of γ significantly Column (5) shows that we obtain an esti-mate of γ = 0041 (std err = 0002) when we measure moversrsquo

27 This coefficient differs slightly from the coefficient of γ = 0038 that weobtain when regressing the coefficients bm on m in Figure IV Panel B becauseestimating the regression in the microdata puts different weights on each age(as we have more data at older ages) while estimating the regression using thebm coefficients puts equal weight on all ages The standard error in this and allsubsequent specifications is also obtained from the regression in the microdataTo simplify computation we report conventional (unclustered) standard errorsClustering standard errors by family to account for correlated outcomes acrosssiblings does not affect the standard errors appreciably In addition regressing theestimates of bm on m in Figure IV which is analogous to clustering the standarderrors by the age at move also yields a standard error of 0002 showing that ourinferences are not sensitive to the way in which standard errors are computed



TA

BL

EII

CH

ILD

HO

OD

EX

PO

SU

RE

EF

FE

CT

ES

TIM

AT

ES

Dep

ende

nt

vari

able

Ch

ildrsquo

sin

com

era

nk

atag

e24

Wit

hfa

mil

yfi

xed

effe

cts

Spe

cifi

cati

on

Poo

led

Age

23

Age

lt18

No

coh

ort

Indi

vidu

alC

hil

dC

ZB

asel

ine

No

coh

ort

Tim

e-co

ntr

ols

inco

me

FE

con

trol

sva

ryin

gco

ntr

ols

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Exp

osu

reef

fect

(γ)

004

00

040

003

70

036

004

10

031

004

40

031

004

3(0

002

)(0

002

)(0

005

)(0

002

)(0

002

)(0

002

)(0

008

)(0

005

)(0

008

)

Nu

mo

fob

s1

553

021

128

777

368

732

31

553

021

155

302

11

473

218

155

302

11

553

021

155

302

1

Not

esT

his

tabl

ere

port

ses

tim

ates

ofan

nu

alch

ildh

ood

expo

sure

effe

cts

onch

ildr

enrsquos

inco

me

ran

ksat

age

24(γ

)T

he

esti

mat

esca

nbe

inte

rpre

ted

asth

eim

pact

ofsp

endi

ng

anad

diti

onal

year

ofch

ildh

ood

ina

CZ

wh

ere

chil

dren

ofpe

rman

ent

resi

den

tsh

ave

1pe

rcen

tile

poin

th

igh

erin

com

era

nks

atag

e24

Sta

nda

rder

rors

are

show

nin

pare

nth

eses

Eac

hco

lum

nre

port

ses

tim

ates

from

are

gres

sion

ofa

chil

drsquos

inco

me

ran

kat

age

24on

the

diff

eren

cebe

twee

npe

rman

ent

resi

den

tsrsquop

redi

cted

ran

ksin

the

dest

inat

ion

vers

us

the

orig

in

inte

ract

edw

ith

the

age

ofth

ech

ild

atth

eti

me

ofth

em

ove

(m)

We

perm

itse

para

teli

nea

rin

tera

ctio

ns

for

m

23an

dm

gt23

and

repo

rtth

eco

effi

cien

ton

the

inte

ract

ion

for

m

23E

ach

regr

essi

onal

soin

clu

des

addi

tion

alco

ntr

ols

spec

ified

ineq

uat

ion

(7)

Per

man

ent

resi

den

tsrsquop

redi

cted

ran

ksar

eco

nst

ruct

edu

sin

gli

nea

rre

gres

sion

sof

chil

dren

rsquosra

nks

onpa

ren

tsrsquor

anks

inea

chC

Zan

dbi

rth

coh

ort

assh

own

inF

igu

reI

Col

um

n(1

)re

port

sth

ees

tim

ate

ofγ

from

equ

atio

n(7

)u

sin

gal

lch

ildr

enin

the

prim

ary

anal

ysis

sam

ple

ofon

e-ti

me

mov

ers

defi

ned

inth

en

otes

toT

able

I(P

anel

C)

Col

um

ns

(2)

and

(3)

rest

rict

the

sam

ple

toth

ose

wh

om

ove

ator

befo

reag

e23

or18

In

colu

mn

(4)

we

excl

ude

the

coh

ort

inte

ract

ion

sw

ith

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

loca

tion

and

inst

ead

incl

ude

asi

ngl

eco

ntr

olfo

rth

epr

edic

ted

outc

omes

ofpe

rman

ent

resi

den

tsin

the

orig

inC

olu

mn

(5)r

epli

cate

sco

lum

n(1

)u

sin

gin

divi

dual

inco

me

ran

ks(r

ath

erth

anh

ouse

hol

din

com

era

nks

)to

mea

sure

both

the

chil

drsquos

outc

ome

and

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

Col

um

n(6

)ad

dsfi

xed

effe

cts

for

the

chil

drsquos

CZ

in20

12to

the

spec

ifica

tion

inco

lum

n(1

)n

ote

that

this

spec

ifica

tion

has

asl

igh

tly

smal

ler

nu

mbe

rof

obse

rvat

ion

sbe

cau

seZ

IPco

dein

form

atio

nis

mis

sin

gfo

rso

me

chil

dren

in20

12(e

g

ach

ild

wit

hn

oea

rnin

gsor

taxa

ble

inco

me)

Col

um

n(7

)ad

dsfa

mil

yfi

xed

effe

cts

toth

eba

seli

ne

spec

ifica

tion

inco

lum

n(1

)C

olu

mn

(8)

adds

fam

ily

fixe

def

fect

sto

the

spec

ifica

tion

inco

lum

n(4

)th

atdo

esn

otin

clu

deco

hor

t-va

ryin

gin

terc

epts

C

olu

mn

(9)

adds

con

trol

sfo

rch

ange

sin

pare

nta

lm

arit

alst

atu

san

din

com

era

nk

inth

eye

arbe

fore

vers

us

afte

rth

em

ove

alon

gw

ith

thei

rin

tera

ctio

ns

wit

hth

eag

eof

the

chil

dat

the

tim

eof

the

mov

ean

din

dica

tors

for

mov

ing

abov

ean

dbe

low

age

23t

oth

esp

ecifi

cati

onin

colu

mn

(7)



income ranks yi and permanent residentsrsquo income ranks ypcs atthe individual rather than household level

We replicate the analysis in Table II at the county level inOnline Appendix Table V We obtain slightly smaller exposureeffect estimates of γ 0035 at the county level indicating thatselection effects account for a larger fraction of the variance inpermanent residentsrsquo outcomes at smaller geographies This isintuitive as families are more likely to sort geographically (eg tobetter school districts) within rather than across labor markets

IVC Mechanisms

In this subsection we present a set of additional specificationsthat shed light on the mechanisms through which neighborhoodsaffect childrenrsquos outcomes

We begin by distinguishing the role of childhood environmentfrom differences caused by variation in labor market conditionsor local costs of living across areas In column (6) of Table II weadd fixed effects for the CZ in which the child lives at age 24(when income is measured) to the baseline model This specifica-tion compares the outcomes of children who live in the same labormarket in adulthood but grew up in different neighborhoods Weobtain an annual exposure effect of γ = 0031 in this specificationindicating that the majority of the exposure effect in our base-line specification is driven by differences in exposure to a betterchildhood environment holding fixed labor market conditions28

This conclusion is consistent with the fact that moving to an areawhere permanent residents have higher income just before enter-ing the labor market (eg in onersquos early twenties) has little effecton income as shown in Figure IV

Next we examine heterogeneity in exposure effects acrosssubsamples (Online Appendix Table III) Standard models oflearning predict that moving to a better area will improve out-comes but moving to a worse area will not In practice the expo-sure effect for negative moves is larger than for positive movesγ = 0030 for moves to better CZs (odps gt 0) while γ = 0040 for

28 This specification likely overadjusts for differences in labor market condi-tions and underestimates γ because the CZ in which the child resides as an adultis itself an endogenous outcome that is likely related to the quality of a childrsquosenvironment For example one of the effects of growing up in a good area may bean increased probability of getting a high-paying job in another city



moves to worse CZs (odps lt 0)29 Spending part of onersquos childhoodin a good neighborhood does not make a child immune to subse-quent deterioration in his or her neighborhood environment Wealso find slightly larger exposure effects for children from above-median income families relative to below-median income families(γ = 0047 versus γ = 0031)

Finally we distinguish between two different mechanismsthat could explain why moving to a better area at a younger ageis more beneficial exposure effectsmdashthe mechanism we have fo-cused on abovemdashand critical age effects Critical age (or criticalperiod) models predict that the impacts of moving to a differentneighborhood vary with childrenrsquos ages (eg Lynch and Smith2005) For example suppose that moving to a better neighbor-hood improves a childrsquos network of friends with a probabilitythat falls with the age at move and that once one makes newcontacts they last forever In this model neighborhood effectswould decline with a childrsquos age at move (as in Figure IV) butthe duration of exposure to a better area would not matter forlong-term outcomes Alternatively if better neighborhoods offera positive treatment (such as better schooling) in each year ofchildhood the key determinant of outcomes would be the totalduration of exposure rather than the specific age at which a childmoves Distinguishing between these mechanisms can be impor-tant for policy the critical age view calls for improving childrenrsquosenvironments at certain key ages while the exposure view callsfor a sustained improvement in environment throughout child-hood

A critical age model cannot be distinguished from an exposureeffect model in a sample of one-time movers because a childrsquos ageat move is perfectly collinear with his or her duration of exposureto the new area However this collinearity is broken when familiesmove multiple times Intuitively one can distinguish between thecritical age and exposure mechanisms by considering children whomove to an area with better permanent residentsrsquo outcomes ypdsbut then move back to the place where they started In this casethe exposure model predicts that children will experience gainsthat are proportional to the number of years they spent in the

29 Moreover roughly an equal fraction of families with children move toCZs with better versus worse outcomes 487 move to CZs with odps gt 0 Thiscontrasts with sorting models suggesting families with children would tend to sortto CZs that produce better outcomes



destination CZ whereas the critical age model predicts that thegain will depend only on the age at which the child first moves tothe new area

To implement this analysis we first generalize the specifica-tion in equation (7) to include families who move more than onceby replacing the odps terms with a duration-weighted measure ofexposure to different areas over childhood (see Online AppendixB for details) This multiple-movers specification yields an annualexposure effect estimate of γ = 0035 (std err = 0001) (OnlineAppendix Table IV column (2)) We test between the critical ageand exposure mechanisms by controlling for the age of the childat the time of each move j interacted with the change in perma-nent residentsrsquo outcomes (od(j)ps) This specification which iso-lates variation in exposure that is orthogonal to the ages at whichchildren move yields an exposure effect estimate of γ = 0034(std err = 0006) (Online Appendix Table IV column (4)) Thesimilarity between these estimates implies that what matters forchildrenrsquos incomes in adulthood is the total time spent in a givenarea (exposure) rather than the age at which one arrives in thatarea30

IVD Summary

Under our key identification assumption (Assumption 1) theempirical results in this section yield three lessons First placematters children who move at earlier ages to areas where priorresidents have higher incomes earn more themselves as adultsSecond place matters in proportion to the duration of childhoodexposure Every year of exposure to a better area during child-hood contributes to higher income in adulthood Third each yearof childhood exposure matters roughly equally The returns togrowing up in a better neighborhood persist well beyond earlychildhood

30 Critical age effects have been most widely documented in linguistic pat-terns and anthropometric measures (eg Singleton and Ryan 2004 Bleakley andChin 2004 van den Berg et al 2014) One potential explanation for why we do notfind evidence of critical age effects here is that we focus on US natives for whomlearning English is presumably less of an issue Neighborhoods do have causaleffects on more nuanced linguistic patterns such as the use of vernacular Englishby African Americans (Rickford et al 2015) Our findings are consistent with suchcultural assimilation mechanisms insofar as they are exposure-dependent



All of these conclusions rest on the assumption that selectioneffects do not vary with the childrsquos age at move We evaluate thevalidity of this assumption in the next section

V VALIDATION OF BASELINE DESIGN

We assess the validity of our key identifying assumptionmdashthat the potential outcomes of children who move to better versusworse areas do not vary with the age at which they movemdashusinga series of tests that focus on different forms of selection andomitted variable bias To organize the analysis we partition theunobserved determinant of childrenrsquos outcomes represented by θ iin equation (3) into two components a component θi that reflectsinputs that are fixed within families such as parent genetics andeducation and a residual component θi = θi minus θi that may varyover time within families such as parentsrsquo jobs

We implement four tests for bias in this section First we ad-dress bias due to selection on fixed family factors θi by comparingsiblingsrsquo outcomes Second we control for changes in parentsrsquo in-come and marital status two key time-varying factors θi that weobserve in our data Our remaining tests focus on unobservabletime-varying factors such as changes in wealth that may havetriggered a move to a better area In our third set of tests weisolate moves that occur due to displacement shocks that inducemany families to move Finally we conduct a set of outcome-basedplacebo (overidentification) tests of the exposure effect model ex-ploiting heterogeneity in permanent residentsrsquo outcomes acrosssubgroups to generate sharp testable predictions about how chil-drenrsquos outcomes should change when they move to different areasIn our view this last approach although least conventional pro-vides the most compelling evidence that the identifying assump-tion holds and that neighborhoods have causal exposure effectson childrenrsquos long-term outcomes

VA Sibling Comparisons

If families with better unobservables (higher θi) move to bet-ter neighborhoods at earlier ages Assumption 1 would be violatedand our estimated exposure effect γ would be biased upward Wecontrol for differences in such family-level factors θi by includingfamily fixed effects when estimating equation (6) For exampleconsider a family that moves to a better area with two children



who are ages m1 and m2 at the time of the move When includ-ing family fixed effects the exposure effect γ is identified by theextent to which the difference in siblingsrsquo outcomes y1 minus y2 co-varies with the difference in their ages interacted with the changein permanent residentsrsquo outcomes (m1 minus m2)odps

Figure V Panel A replicates Figure IV Panel B adding familyfixed effects to equation (6) The linear decline in the estimatedvalues of bm until age 23 is very similar to that in the baselinespecification Children who move to a better area at younger ageshave better outcomes than do their older siblings Regressing thebm coefficients on m for m 23 yields an average annual exposureeffect estimate of γ = 0043 (std err = 0003) very similar to ourestimates above

The selection effect (ie the level of bm after age 24) falls fromδ = 023 in the baseline specification to δ = 001 (not significantlydifferent from 0) with family fixed effects31 Family fixed effectsthus reduce the level of the bm coefficients by accounting for dif-ferential selection in which types of families move to better versusworse areas but do not affect the slope of the bm coefficients Thisis precisely what we should expect if selection effects in wherefamilies choose to move do not vary with childrenrsquos ages whenthey move as required by Assumption 1

Table II Column (7) shows that adding family fixed effects tothe linear specification in equation (7) and estimating the modeldirectly on the micro data yields an estimate of γ = 0044 Othervariants of this regression specification analogous to those incolumns (2)ndash(6) of Table II all yield very similar estimates of γ with one exception excluding the cohort interactions with yposand odps as in column (4) yields γ = 0031 (Table II column(8)) The reason that the estimate of γ falls in this specificationis that we observe childrenrsquos origin locations for fewer years inearlier birth cohorts as discussed in Section IVA The missingdata on origins increases the level of the selection effect δ in ear-lier cohorts (see Online Appendix A) Because we only observemoves at older ages for children in earlier cohorts these differ-ences across cohorts induce a positive correlation between δm andm biasing our estimate of γ downward This bias is magnified inthe specifications with family fixed effects because they are iden-tified purely by comparing the outcomes of children in different

31 δ is identified even with family fixed effects because odps varies acrossbirth cohorts



FIGURE V

Childhood Exposure Effects on Income Ranks with Additional Controls

This figure replicates Figure IV Panel B using specifications analogous to equa-tion (6) that include family fixed effects (Panel A) and family fixed effects andcontrols for changes in marital status and parental income around the time of themove (Panel B) To control for changes in parental income we construct parentalincome ranks by childrsquos birth cohort and calendar year We interact the differencesin parental ranks in the year before versus after the move with the childrsquos age atthe time of the move along with interactions with indicators for moving beforeversus after age 23 To control for changes in marital status we construct indica-tors for being always married getting divorced or being never married in the yearbefore the move and the year after the move (getting married is the omitted cate-gory) We then interact these marital status indicators with the childrsquos age at thetime of the move along with interactions with indicators for moving above versusbelow age 23 See notes to Figure IV for additional details on the construction ofthe figure



birth cohorts whereas our baseline specifications also comparechildren in the same birth cohort whose parents move at differ-ent times Including cohort interactions with odps eliminates thisbias by permitting a separate selection term δ for each cohort32

In sum we continue to find childhood exposure effects of γ 004 when comparing siblingsrsquo outcomes implying that our designis not confounded by differences in the types of families who moveto better areas when their children are younger

VB Controls for Time-Varying Observables

The research design in Figure V Panel A accounts for biasdue to fixed differences in family inputs θi but it does not accountfor time-varying inputs θi For example moves to better areas maybe triggered by events such as job promotions that directly affectchildrenrsquos outcomes in proportion to their time of exposure to thedestination Such shocks could bias our estimate of β upward evenwith family fixed effects

Prior research has focused on changes in parentsrsquo income andmarital status as two key factors that may induce moves and alsodirectly affect childrenrsquos outcomes in adulthood (eg Jencks andMayer 1990) We can directly control for these two time-varyingfactors in our data as we observe parentsrsquo incomes and maritalstatus in each year from 1996 to 2012 We control for the effects ofchanges in income around the move when estimating equation (6)by including controls for the change in the parentrsquos income rankfrom the year before to the year after the move interacted withindicators for the childrsquos age at move The interactions with age atmove permit the effects of income changes to vary with the dura-tion of childhood exposure to higher versus lower levels of parentincome Similarly we control for the impact of changes in mari-tal status by interacting indicators for each of the four possiblechanges in the motherrsquos marital status in the year before versusafter the move (married to unmarried unmarried to married un-married to unmarried and married to married) with indicatorsfor the childrsquos age at move

Figure V Panel B replicates Panel A controlling for all ofthese variables in addition to family fixed effects Controlling forchanges in parent income and marital status has little effect on

32 The attenuation bias in γ is further amplified in CZs with smaller popu-lations where odps is measured with greater error (see Online Appendix A andAppendix Table VI)



the estimates of bm The estimates of γ = 0042 and δ = 0015are virtually identical to those when we do not control for thesetime-varying factors Table II column (9) confirms that includingthese controls in a linear regression estimated on the micro datayields similar estimates

These results show that changes in income and family struc-ture are not a significant source of bias in our design Howeverother unobserved factors could still be correlated with moving to abetter or worse area in a manner that generates omitted variablebias The fundamental identification problem is that any unob-served shock that induces child irsquos family to move to a differentarea could be correlated with parental inputs θ i These changesin parental inputs could potentially increase the childrsquos income yiin proportion to the time spent in the new area even in the ab-sence of neighborhood effects For example a wealth shock mightlead a family to both move to a better neighborhood and increaseinvestments in the child in the years after the shock which couldimprove yi in proportion to exposure time independent of neigh-borhood effects In the next two subsections we address concernsabout bias due to such unobserved factors

VC Displacement Shocks

One approach to accounting for unobservable shocks is toidentify moves where we have some information about the shockthat precipitated the move Suppose we identify families who wereforced to move from an origin o to a nearby destination d because ofan exogenous shock such as a natural disaster Such displacementshocks can induce differential changes in neighborhood quality asmeasured by permanent residentsrsquo outcomes (odps) For instanceHurricane Katrina displaced families from New Orleans (an areawith relatively poor outcomes compared with surrounding areas)leading to an increase in average neighborhood quality for dis-placed families (odps gt 0) In contrast Hurricane Rita hit Hous-ton an area with relatively good outcomes and may have reducedneighborhood quality (odps lt 0) If these displacement shocks donot have direct exposure effects on children that are correlatedwith odpsmdashfor example the direct effects of the disruption in-duced by hurricanes does not covary with neighborhood qualitychangesmdashthen Assumption 1 is satisfied and we obtain unbiasedestimates of the exposure effect γ by focusing on displaced fami-lies Conceptually by isolating a subset of moves caused by known



exogenous shocks we can more credibly ensure that changes inchildrenrsquos outcomes are not driven by unobservable factors33

To operationalize this approach we first identify displace-ment shocks based on population outflows at the ZIP code levelLet Kzt denote the number of families who leave ZIP code z in yeart in our sample of one-time movers and Kz denote mean outflowsbetween 1996 and 2012 We define the shock to outflows in year tin ZIP z as kzt = Kzt

Kz34

Though many of the families who move in subsamples withlarge values of kzt do so for exogenous reasons their destination dis still an endogenous choice that could lead to bias For examplefamilies who choose to move to better areas (higher ypds) wheninduced to move by an exogenous shock might also invest morein their children To reduce potential biases arising from the en-dogenous choice of destinations we isolate variation arising fromthe average change in neighborhood quality for individuals whoare displaced Let E[odps|q z] denote the difference in the meanpredicted outcome in the destination CZs relative to the origin CZfor individuals in origin ZIP code z and parent income decile q(averaging over all years in the sample not just the year of theshock) We instrument for the difference in predicted outcomes ineach familyrsquos destination relative to origin (odps) with E[odps|qz] and estimate the linear specification in equation (7) using 2SLSto identify the exposure effect γ IV35

Figure VI presents the results of this analysis To constructthis figure we take ZIP-year cells with above-median outflows

33 This research design is related to Sacerdotersquos (2012) analysis of the effectsof Hurricanes Katrina and Rita on student test score achievement Although weuse similar variation we do not focus on the direct effects of the displacementitself but on how childrenrsquos long-term outcomes vary in relation to the outcomesof permanent residents in the destination to which they were displaced

34 Searches of historical newspaper records for cases with the highest outflowrates kzt reveal that they are frequently associated with events such as natural dis-asters or local plant closures Unfortunately there is insufficient power to estimateexposure effects purely from the events identified in newspapers

35 This approach does not fully eliminate the scope for selection bias asbiases from the endogenous choice of destinations could persist if there is unob-served heterogeneity across areas experiencing displacement shocks However itreduces the scope for selection bias by focusing on moves induced by aggregatedisplacement shocks and eliminating variation in odps due to individual choicewhich is more likely to be correlated with unobservables θ i than the area-levelvariation in E[odps|q z] By testing if the estimate of γ remains stable when weuse an estimator that reduces the scope for selection we can gauge whether ourbaseline estimate of γ is biased



FIGURE VI

Exposure Effect Estimates Using Displacement Shocks

This figure presents estimates of annual childhood exposure effects (γ ) for thesubset of areas that experience displacement shocks defined as ZIP code by yearcells that have large outflows in the number of residents We measure outflowsby defining Kzt as the number of families who leave ZIP code z in year t in ourone-time movers sample and Kz as mean outflows between 1996 and 2012 Wedefine the shock to outflows in year t in ZIP z as kzt = Kzt

Kz We then take ZIP-year

cells with above-median outflows (kzt gt 117) and divide them into 25 population-weighted bins based on the size of the shock kzt For each subset of observationswith values of kzt above the percentile threshold listed on the x-axis we estimateγ using equation (7) instrumenting for the change in predicted outcomes basedon permanent residents odps with the average change in predicted outcomes formovers from the origin ZIP E[odps|q z] We define E[odps|q z] as the meanvalue of odps for each parental income decile q pooling across all years and allmovers out of ZIP code z The figure plots the resulting estimates of γ versus thepercentile threshold cutoff for the sample The dashed lines show 95 confidenceintervals for the estimates The mean value of the outflow shock kzt used in eachsubsample is shown in brackets below the percentile thresholds

(kzt gt 117) and divide them into population-weighted bins basedon the size of the shock kzt36 The first point in Figure VI shows the2SLS estimate of the annual exposure effect γ IV using all obser-vations with kzt greater than its median value (117) The secondpoint shows the estimate of γ IV using all observations with kzt at orabove the 52nd percentile The remaining points are constructedin the same way increasing the threshold by two percentiles ateach point with the last point representing an estimate of γ IV

36 To ensure that large outflows are not driven by areas with small pop-ulations we exclude ZIP-year cells with fewer than 10 children leaving in thatyear



using data only from ZIP codes in the highest two percentiles ofoutflow rates The dotted lines show a 95 confidence interval forthe regression coefficients

If the baseline estimates were driven entirely by selectionγ IV would fall to 0 as we limit the sample to individuals who aremore likely to have been induced to move because of an exogenousdisplacement shock But the coefficients remain quite stable atγ IV 004 even when we restrict to moves that occurred as part oflarge displacements That is when we focus on families who moveto a better area for what are likely to be exogenous reasons wecontinue to find that children who are younger at the time of themove earn more as adults

These findings support the view that our baseline estimatesof exposure effects capture the causal effects of neighborhoodsrather than other unobserved factors that change when familiesmove Moreover they indicate that the treatment effects of movingto a different area are similar for families who choose to move foridiosyncratic reasons and families who are exogenously displacedThis result suggests that the exposure effects identified by ourbaseline design can be generalized to a broader set of familiesbeyond those who choose to make a particular move

VD Outcome-Based Placebo Tests

As a final approach to test for bias due to unobservable fac-tors we implement placebo tests that exploit the heterogeneityin permanent residentsrsquo outcomes across subgroups We exploitvariation along three dimensions birth cohorts quantiles of theincome distribution and child gender The causal exposure effectmodel predicts precise convergence of a childrsquos outcome to perma-nent residentsrsquo outcomes for his or her own subgroup In contrastwe argue below that omitted variable and selection models wouldnot generate such subgroup-specific convergence under plausibleassumptions about parentsrsquo information sets and preferences Theheterogeneity in permanent residentsrsquo outcomes thus gives us arich set of overidentifying restrictions to test whether neighbor-hoods have causal effects37 We consider each of the three dimen-sions of heterogeneity in turn

1 Birth Cohorts Although permanent residentsrsquo outcomes aregenerally very stable over time outcomes in some areas (such as

37 In addition to being useful for identification these results are also of directinterest in understanding the heterogeneity of place effects across subgroups



Oklahoma City OK) have improved over time while others (suchas Sacramento CA) have gotten worse38 Such changes could oc-cur for instance because of changes in the quality of local schoolsor other area-level characteristics that affect childrenrsquos outcomesWe exploit this heterogeneity across birth cohorts to test for con-founds in our baseline research design

Under the causal exposure effect model when a childrsquos fam-ily moves to destination d the difference in permanent residentsrsquooutcomes odp s(i) for that childrsquos own birth cohort s(i) should pre-dict his or her outcomes more strongly than the difference in out-comes odps for other cohorts s = s(i) Intuitively what matters fora childrsquos outcome is a neighborhoodrsquos quality for his own cohortnot the neighborhoodrsquos quality for younger or older cohorts In con-trast it is unlikely that other unobservables θ i will vary sharplyacross birth cohorts s in association with odps because the fluctu-ations across birth cohorts are realized only in adulthood and thuscannot be directly observed at the time of the move39 Thereforeby testing whether exposure effects are predicted by a childrsquos ownversus surrounding cohorts we can assess the importance of biasdue to unobservables

We implement this analysis by estimating the baseline spec-ification in equation (7) replacing the change in permanent res-identsrsquo outcomes for the childrsquos own cohort odp s(i) with anal-ogous predictions for adjacent birth cohorts s(i) + t odp s(i) + t(see Online Appendix D for details) The series in red trianglesin Figure VII (color version online) plots the exposure effect es-timates (γt) obtained from these regressions with t ranging fromminus4 to 4 The estimates of γt are similar to our baseline estimateof γ = 0040 for the leads and lags consistent with the high de-gree of serial correlation in permanent residentsrsquo outcomes Theseries in blue circles plots analogous coefficients γt when all thecohort-specific predictions from the four years before to the fouryears after the childrsquos own cohort are included simultaneouslyIn this specification the coefficients on the placebo exposure ef-fects (γt for t = 0) are all very close to 0mdash and not statistically

38 The autocorrelation of ypcs with ypcsminus1 across childrenrsquos birth cohorts is095 at the 25th percentile of the parent income distribution

39 For instance a family that moves with a 10-year-old child will not observeypds for another 14 years (if income is measured at age 24)



FIGURE VII

Exposure Effect Estimates Based on Cross-Cohort Variation

This figure presents estimates of the annual childhood exposure effect on chil-drenrsquos income ranks in adulthood using permanent resident predictions for thechildrsquos own birth cohort and surrounding ldquoplacebordquo birth cohorts The series intriangles plots estimates of the exposure effect γ t from nine separate regressionsanalogous to that in equation (7) using permanent resident predictions from co-hort s + t (where t ranges between minus4 and 4) as the key independent variablesand the outcomes of children in birth cohort s as the dependent variable By con-struction the exposure effect estimate for t = 0 (highlighted by the dashed verticalline) corresponds to the baseline estimate of γ = 0040 in column (1) of Table IIThe series in circles plots estimates from a single multivariable regression thatsimultaneously includes all nine permanent resident predictions t = minus4 4 andplots the coefficient on the interaction of the childrsquos age at the time of the movem with odp s+t the difference between permanent residentsrsquo predicted ranks inthe destination versus the origin in cohort s + t The figure also reports p-valuesfrom two hypothesis tests the hypothesis that γ (the estimate using the actualcohort t = 0) equals 0 in the simultaneous specification and the hypothesis thatall other coefficients γ s+t excluding the own-cohort coefficient are equal to 0 SeeOnline Appendix D for further details on the regression specifications

significant40 However the exposure effect estimate for the childrsquosown cohort remains at approximately γ = 004 even when we con-trol for the surrounding cohortsrsquo predictions and is significantlydifferent from the estimates of γt for t = 0 (p lt 001)

The evidence in Figure VII strongly supports the view that thechange in childrenrsquos outcomes is driven by causal effects of expo-sure to a different place Intuitively it is unlikely that a correlated

40 A test of the joint hypothesis that all γt = 0 for all t = 0 yields a p-value of251



shock (such as a change in wealth when the family moves) wouldcovary precisely with cohort-level differences in place effects asmanifested in the outcomes of children of permanent residentsFormally this test relies on the assumption that if unobservablesθ i are correlated with exposure to a given cohort s(i)rsquos place ef-fect (proxied for by permanent residentsrsquo outcomes) they mustalso be correlated with exposure to the place effects of adjacentcohorts t

(8) Cov(θi modps(i)|X) gt 0 rArr Cov(θi modpt|X modps(i)) gt 0

where X represents the vector of fixed effects and other controlsin equation (7) Under this assumption the findings in Figure VIIimply that our estimates of γ reflect causal neighborhood effects(which are cohort-specific) rather than omitted variables whichare not cohort-specific under equation (8)

2 Quantiles Distributional Convergence Places differ notonly in childrenrsquos mean outcomes but also in the distribution ofchildrenrsquos outcomes For example children who grow up in low-income families in Boston and San Francisco have comparablemean ranks but children in San Francisco are more likely to endup in the tails of the income distribution than those in Boston Ifneighborhoods have causal exposure effects we would expect con-vergence in moversrsquo outcomes not just at the mean but across theentire distribution in proportion to exposure time In contrast it isless plausible that omitted variables such as wealth shocks wouldperfectly replicate the distribution of outcomes of permanent res-idents in each CZ41 Therefore testing for quantile-specific con-vergence can distinguish the causal exposure effect model fromomitted variable explanations

To implement these tests we begin by constructing predic-tions of the probability of having an income in the upper or lowertail of the national income distribution at age 24 for children ofpermanent residents in each CZ c In each CZ we regress an in-dicator for a child being in the top 10 of the distribution or anindicator for not being employed on parent income rank p using

41 Families are unlikely to be able to forecast their childrsquos eventual quantilein the income distribution making it difficult to sort precisely on quantile-specificneighborhood effects Even with such knowledge there is no ex ante reason toexpect unobserved shocks such as changes in wealth to have differential andpotentially nonmonotonic effects across quantiles in precise proportion to theoutcomes in the destination



TABLE IIICHILDHOOD EXPOSURE EFFECT ESTIMATES DISTRIBUTIONAL CONVERGENCE

Dep var Upper-tail child in top decile Lower-tail child not employed

(1) (2) (3) (4) (5) (6)

Distributional prediction 0043 0040 0041 0043(0002) (0003) (0003) (0003)

Mean rank prediction 0024 0003 0018 minus0002(placebo) (0002) (0003) (0002) (0003)

Num of obs 1553021 1553021 1553021 1553021 1553021 1553021

Notes This table reports estimates of annual childhood exposure effects (γ ) for upper-tail and lower-tailoutcomes being in the top 10 of the cohort-specific income distribution at age 24 or not being employedStandard errors are shown in parentheses Column (1) reports estimates from a regression of an indicator forbeing in the top 10 on the difference between permanent residentsrsquo predicted probabilities of being in theupper-tail in the destination versus the origin interacted with the age of the child at the time of the move (m)We permit separate linear interactions for m 23 and m gt 23 and report the coefficient on the interactionfor m 23 The regression also includes additional controls analogous to those in equation (7) see OnlineAppendix D for details Column (2) continues to use the indicator for being in the top 10 as the dependentvariable but uses the difference between permanent residentsrsquo predicted mean ranks in the destinationversus the origin instead of their predicted upper-tail probabilities on the right side of the regression Column(3) includes both the upper-tail (distributional) prediction as well as the mean rank prediction in the sameregression Columns (4)ndash(6) replicate columns (1)ndash(3) using an indicator for being unemployed at age 24 as theoutcome and the difference between the permanent residentsrsquo predicted probabilities of being unemployed inthe destination versus the origin as the key independent variable Employment is defined as an indicator forhaving a W-2 filed on onersquos behalf at age 24 In all columns the sample consists of all children in the primaryanalysis sample of one-time movers defined in the notes to Table I (Panel C)

an equation analogous to equation (1) including a quadratic termin parental income rank p to account for the nonlinearities in tailoutcomes identified in Chetty et al (2014) We then calculate thepredicted probability of being nonemployed πU

pcs and being abovethe 90th percentile π90

pcs using the fitted values from these regres-sions as in equation (2)

In Table III we estimate exposure effect models analogousto equation (7) using these distributional predictions instead ofmean predictions In columns (1)ndash(3) the dependent variable isan indicator for having income in the top 10 of the income dis-tribution Column (1) replicates the baseline specification in equa-tion (7) using 90

odps = π90pds minus π90

pos instead of the mean predictionodps = ypds minus ypos as the key independent variable (see OnlineAppendix D for the exact regression specifications) We obtainan exposure effect estimate of γ = 0043 a year in this specifi-cation Column (2) uses the change in the predicted mean rankodps instead Here we obtain a statistically significant estimateof 0024 as expected given the high degree of correlation in per-manent residentsrsquo outcomes across quantiles places where morechildren reach the top 10 also tend to have higher mean incomesIn column (3) we include both the quantile prediction 90

odps and



the mean prediction odps identifying the coefficients purely fromdifferential variation across quantiles within CZs The coefficienton the quantile prediction remains unchanged at approximatelyγ = 004 while the coefficient on the mean prediction is not sig-nificantly different from 0

Table III columns (4)ndash(6) replicate columns (1)ndash(3) using anindicator for nonemployment as the dependent variable and theprediction for nonemployment U

odps instead of 90odps as the key in-

dependent variable As in the upper tail childrenrsquos probabilitiesof being in the lower tail of the income distribution are fully de-termined by the quantile-specific prediction rather than the meanprediction In column (6) the coefficient on the nonemploymentprediction U

odps is γ = 0043 and the placebo coefficient on themean rank prediction is minus0002

In short we find evidence of distributional convergence con-trolling for mean incomes the distribution of childrenrsquos incomesconverges to the distribution of incomes in the destination in pro-portion to exposure time at a rate of approximately 4 a year42

Because omitted variables such as wealth shocks would be un-likely to generate such distributional convergence this findingagain supports the view that the convergence in moversrsquo outcomesis driven by causal effects of place Formally assume that if unob-servables θ i are correlated positively with exposure to place effectson upper (or lower) tail outcomes π

qpcs they must also be correlated

with exposure to the place effects on mean incomes (proxied forby permanent residentsrsquo outcomes)

(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0

Under this assumption the findings in Table III imply that ourestimates of γ reflect causal place effects (which are quantile-specific) rather than omitted variables which are not quantile-specific under equation (9)

3 Gender Finally we conduct an analogous set of placebotests exploiting heterogeneity in permanent residentsrsquo out-comes by child gender We begin by constructing gender-specific

42 The rate of convergence need not be identical across all quantiles ofthe income distribution because the prediction for permanent residents at eachquantile π90

pcs could reflect a different combination of causal effects and sortingThe key test is whether the prediction for the relevant quantile has more predic-tive power than predictions at the mean or other quantiles



TABLE IVCHILDHOOD EXPOSURE EFFECT ESTIMATES GENDER-SPECIFIC CONVERGENCE

Dependent variable childrsquos income rank at age 24

No family fixed effects With family fixed effects

(1) (2) (3) (4) (5) (6) (7)

Own gender 0038 0030 0031 0027 0030prediction (0002) (0003) (0006) (0006) (0007)

Other gender 0031 0010 0016 0017 0009prediction (placebo) (0002) (0003) (0005) (0005) (0007)

Sample Full sample Full sample 2-gender HH

Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964

Notes This table reports estimates of annual childhood exposure effects (γ ) using gender-specific permanentresident predictions Standard errors are shown in parentheses In all columns the dependent variable is thechildrsquos family income rank at age 24 In columns (1)ndash(6) the sample consists of all children in the primaryanalysis sample of one-time movers defined in the notes to Table I (Panel C) Column (1) replicates column (1)of Table II replacing the predicted outcomes based on all permanent residents in the origin and destinationwith predictions based on the outcomes of children who have the same gender as the child who movesColumn (2) replicates column (1) replacing the own-gender predicted outcomes with the predicted outcomesof the opposite gender Column (3) combines the variables in columns (1) and (2) including both the own-gender and other-gender (placebo) predictions Columns (4)ndash(6) replicate (1)ndash(3) including family fixed effectsColumn (7) replicates column (6) restricting the sample of movers to families with at least one child ofeach gender Each regression also includes additional controls analogous to those in equation (7) see OnlineAppendix D for details

predictions of the mean household income ranks of children of per-manent residents by estimating equation (1) separately for maleand female children which we denote by ym

pcs and y fpcs Places

that are better for boys are generally better for girls as well the(population-weighted) correlation of ym

pcs and y fpcs across CZs is

093 at the median (p = 50)43 We exploit the residual variationacross genders to conduct placebo tests analogous to those abovebased on the premise that unobservable shocks are unlikely tohave gender-specific effects

In Table IV we estimate exposure effect models analogousto equation (7) with separate predictions by gender Column (1)replicates equation (7) using the gender-specific prediction

godps

instead of the prediction that pools both genders We obtain anexposure effect estimate of γ = 0038 per year in this specifica-tion In column (2) we use the prediction for the other gender

minusgodps instead Here we obtain an estimate of 0031 as expected

given the high degree of correlation in outcomes across genders

43 Online Appendix Figure V presents choropleth maps of ympcs minus y f

pcs at p = 25and p = 75 For low-income families (p = 25) outcomes for boys are relatively worsethan those for girls in areas with higher crime rates a larger fraction of singleparents and greater inequality (Chetty et al 2016)



In column (3) we include predictions for both genders identifyingthe coefficients purely from differential variation across genderswithin CZs In this specification the coefficient on the own genderprediction is γ = 003 three times larger than the other-genderprediction which is close to 044

One may be concerned that families sort to different areasbased on their childrsquos gender whichmdashunlike the quantile andcohort-specific variation used abovemdashis known at the time of themove To address this concern Table IV columns (4)ndash(6) replicatecolumns (1)ndash(3) including family fixed effects The own-gender pre-diction remains a stronger predictor of childrenrsquos outcomes thanthe other-gender prediction even when we compare siblingsrsquo out-comes within families Column (7) shows that this remains thecase when we restrict the sample to families that have at leastone boy and one girl for whom differential sorting by gender isinfeasible

The gender-specific convergence documented in Table IV sup-ports the causal exposure effects model under an assumption anal-ogous to equation (8) namely that the unobservable θ i does notvary differentially across children of different genders within afamily This assumption requires that families who move to ar-eas that are particularly good for boys do not systematically in-vest more in their sons relative to their daughters a restrictionthat would hold if for instance families do not have differentpreferences over their sonsrsquo and daughtersrsquo outcomes Under thisassumption the gender-specific convergence in proportion to ex-posure time must reflect causal place effects

VE Summary

The results in this section show that various refinements ofour baseline designmdashsuch as including family fixed effects or ex-ploiting cohort- or gender-specific variationmdashall yield annual ex-posure effect estimates of γ 004 These findings imply that anyomitted variable θ i that generates bias in our estimate of the ex-posure effect γ must (i) operate within families in proportion toexposure time (family fixed effects) (ii) be orthogonal to changes

44 It is not surprising that the other gender prediction remains positive asthe prediction for the other gender may be informative about a placersquos effect forchildren of a given gender due to measurement error In general finding a 0 effecton the ldquoplacebordquo prediction is sufficient but not necessary to conclude that there isno sorting under an assumption analogous to equation (8)



in parental income and marital status (controls for observables)(iii) persist in the presence of moves induced by displacementshocks (displacement shock analysis) and (iv) precisely replicatepermanent residentsrsquo outcomes by birth cohort quantile and gen-der in proportion to exposure time (outcome-based placebo tests)We believe that plausible omitted variables are unlikely to haveall of these properties and therefore conclude that our estimate ofγ 004 is an unbiased estimate of the annual childhood exposureeffect

The convergence rate of 4 per year of exposure between theages 9 to 23 implies that children who move at age 9 would pick upabout (23 minus 9) times 4 = 56 of the observed difference in permanentresidentsrsquo outcomes between their origin and destination CZs Ifwe assume that the rate of convergence remains at 4 even beforeage 9 our estimates would imply that children who move at birthto a better area and stay there for 20 years would pick up about80 of the difference in permanent residentsrsquo outcomes betweentheir origins and destinations

An auxiliary implication of the results in this section is thatthe simple baseline design of comparing families who move withchildren of different ages is not confounded by selection andomitted variable biases Although there is clear evidence of selec-tion in terms of where families movemdashas shown by the estimate ofδ gt 0 in Figure IV mdash we find no evidence of differential selectionbased on when families move to a better versus worse area (at leastafter their children are nine years old)45 This finding implies thatresearch designs exploiting variation in the timing of moves canbe used to identify the causal effects of neighborhoods in observa-tional data providing a scalable tool for identifying neighborhoodeffects even in the absence of randomized experiments

VI OTHER OUTCOMES

In this section we estimate neighborhood effects for sev-eral other outcomes beyond income college attendance marriageteenage birth and teenage employment This analysis provides

45 Such differential selection might be small because the outcomes of childrenof permanent residents ypcs are not highly correlated with mean parent incomesacross areas (Chetty et al 2014) As a result moving to a better area for children(higher ypcs) is not systematically associated with parents finding higher-payingjobs mitigating what might be the most important confounding factor for ourdesign



further evidence on the types of outcomes that are shaped byneighborhoods and illustrates how neighborhoods affect behaviorbefore children enter the labor market

Figure VIII replicates Figure IV Panel B using college at-tendance and marriage as the outcomes In Panel A we repli-cate the specification in equation (6) replacing odps with C

odps =Cpds minus Cpos where Cpcs is the fraction of children who attend col-lege at any point between ages 18 and 23 (among children of per-manent residents in CZ c in birth cohort s with parental incomerank p) In Panel B we replace odps with M

odps = Mpds minus Mposwhere Mpcs is the fraction of children who are married at age 26

We find evidence of childhood exposure effects until age 23for both of these outcomes Moving to an area with higher collegeattendance rates at a younger age increases a childrsquos probabilityof attending college Likewise moving at a younger age to anarea where permanent residents are more likely to be marriedincreases a childrsquos probability of being married Using parametricmodels analogous to equation (7) the estimated annual exposureeffect for college attendance is comparable to our estimates forincome (γ = 0037) and is slightly smaller for marriage (γ = 0025)

In Figure VIII Panels C and D we analyze outcomes mea-sured while children are teenagers Panel C considers teen birthdefined (for both men and women) as having a child between theages of 13ndash19 We construct gender-specific predictions of teenagebirth rates and plot estimates from the baseline specification inequation (6) replacing odps with z

odpsg = zpdsg minus zposg where zpcsgis the fraction of children of permanent residents with parentalincome p in CZ c cohort s and gender g who have a teenage birthFor both boys and girls there are clear childhood exposure effectsmoving at an earlier age to an area with a higher teen birth rateincreases a childrsquos probability of having a teenage birth The gra-dient is especially steep between ages 13 and 18 suggesting thata childrsquos neighborhood environment during adolescence may playa particularly important role in determining teen birth outcomes

In Figure VIII Panel D we analyze neighborhood effects onteenage employment rates measured at age 16 The key inde-pendent variable (corresponding to odps) in this figure is thedifference in age 16 employment rates of children of permanentresidents in the destination versus the origin CZ We find a dis-continuous effect of moving just before age 16 on the probabilityof working at 16 rather than a continuous exposure effect Chil-dren who move at age 15 to a CZ where more 16-year-olds work



FIGURE VIII

Exposure Effects on College Attendance Marriage Teen Birth and TeenEmployment

This figure plots exposure effects for college marriage and the outcomes ofteenagers using an approach analogous to that in Figure IV Panel B In PanelA we replicate the specification in equation (6) using an indicator for college at-tendance at any age between 18 and 23 as the dependent variable instead of thechildrsquos income rank and replacing the key independent variable odps = ypds minus yposwith the difference between permanent residentsrsquo college attendance rates inthe destination versus the origin The coefficients that are plotted can thereforebe interpreted as the effect of moving to an area where permanent residentsrsquocollege attendance rates are 1 percentage point higher at age m We require thatthe child be observed between ages 18 and 23 to define college attendance be-cause we observe college attendance in years 1999ndash2012 we obtain estimatesfor children who move between the ages of 8 and 29 In Panel B we replicatethe baseline specification in equation (6) replacing the childrsquos outcomes with anindicator for being married at age 26 and replacing odps = ypds minus ypos with thedifference between permanent residentsrsquo marriage rates in the destination versusthe origin Panel C replicates the parametric specification in equation (6) usingteenage birth as the dependent variable and replacing the key independent vari-able odps = ypds minus ypos with the difference between permanent residentsrsquo teenbirth rates in the destination versus the origin We define teenage birth as havinga child between the ages of 13 and 19 using data from the Social Security Ad-ministrationrsquos DM-2 database and estimate separate specifications for males andfemales who have a child Panel D replicates the parametric specification in equa-tion (6) using an indicator for working at age 16 (based on having a W-2) as thedependent variable and replacing the key independent variable odps = ypds minus yposwith the difference between permanent residentsrsquo teen employment rates in thedestination versus the origin at the corresponding age The coefficients that areplotted can therefore be interpreted as the effect of moving at age m to an areawhere permanent residentsrsquo teen employment rates are 1 percentage point higherat age 16 Age 16 is shown by the vertical dashed line because moves after theage at which employment is measured cannot have a causal effect the coefficientsto the right of the dashed lines reflect a selection effect See notes to Figure IV forfurther details on the construction of this figure



are much more likely to work at age 16 than children who makethe same move at age 17 Making the same move at earlier ages(before age 16) further increases the probability of working atage 16 but the exposure effect is small relative to the jump atage 16 itself46 Analogous jumps are observed at ages 17 and 18when we measure employment at ages 17 and 18 (Online Ap-pendix Figure VI) These jumps suggest that neighborhood effectsmay be partly driven by distinct experiences at different points ofchildhood such as summer jobs that are available in a given areaat certain ages Such age-specific impacts may aggregate to pro-duce the linear childhood exposure effects that shape outcomes inadulthood

Although the mean income of individuals in an area is corre-lated with other outcomes such as college attendance and teenagebirth rates there is substantial independent variation in eachof these outcomes For example permanent residentsrsquo mean in-come ranks at age 30 have a (population-weighted) correlationof 046 with college attendance rates for children with parentsat p = 25 (Online Appendix Table VII) Hence the finding thatmoversrsquo outcomes converge to those of permanent residents on allof these dimensions constitutes further evidence that neighbor-hoods have causal effects as it would be unlikely that unobservedconfounds would generate such convergence on a spectrum of dif-ferent outcomes47 Moreover the fact that neighborhoods havecausal effects on a wide variety of outcomes beyond earnings fur-ther suggests that the mechanism through which neighborhoodsshape childrenrsquos outcomes is not driven by labor market conditionsbut rather a set of environmental factors that shape behaviorsthroughout childhood

VII CONCLUSION

This article has shown that childrenrsquos opportunities for eco-nomic mobility are shaped by the neighborhoods in which theygrow up Neighborhoods affect childrenrsquos long-term outcomes

46 The magnitudes of the bm coefficients in Panel D are approximately 08at young ages and 0 after age 16 Under our identifying assumption of constantselection effects by age this implies that children who move at birth pick up 80of the differences in teenage employment rates across CZs observed for permanentresidents

47 This logic is analogous to the tests for distributional convergence inSection VD here we effectively test for convergence in the joint distribution ofincome and various other outcomes



through childhood exposure effects every extra year a childspends growing up in an area where permanent residentsrsquo incomesare higher increases his or her income Moversrsquo outcomes convergeto those of permanent residents in the destination to which theymove at a rate of approximately 4 per year of childhood exposure(up to age 23) This estimate implies that much of the variationin intergenerational mobility observed across CZs and countiesis driven by causal effects of place rather than differences in thetypes of people living in those places For instance children whomove at age 9 (the earliest age we observe in our data) wouldpick up 56 of the observed difference in permanent residentsrsquooutcomes between their origin and destination CZs If the rate ofconvergence remains at 4 even before age 9mdasha strong assump-tion that should be evaluated in future workmdashour estimates implythat children who move at birth to a better area and stay there for20 years would pick up about 80 of the difference in permanentresidentsrsquo outcomes between their origins and destinations

These results motivate place-focused approaches to improv-ing economic mobility such as making investments to improveoutcomes in areas that currently have low levels of mobility orhelping families move to higher opportunity areas Identifyingspecific policy solutionsmdashthat is the investments needed to im-prove mobility and the areas to which families should be en-couraged to movemdashrequires identifying the causal effect of eachneighborhood and understanding what makes some areas producebetter outcomes than others The analysis in the present articleshows that differences in permanent residentsrsquo outcomes are pre-dictive of neighborhoodsrsquo causal effects on average However itdoes not provide estimates of the causal effect of each area onchildrenrsquos outcomes as the outcomes of permanent residents inany given area will reflect a different mix of selection and causaleffects We construct estimates of the causal effect of growing upin each CZ and county in the United States and characterize theproperties of areas that produce good outcomes in the next articlein this series

STANFORD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RE-SEARCH

HARVARD UNIVERSITY AND NATIONAL BUREAU OF ECONOMIC RE-SEARCH



SUPPLEMENTARY MATERIAL

An Online Appendix for this article can be found at The Quar-terly Journal of Economics Code used to generate tables and fig-ures in this article can be found in Chetty and Hendren (2018b)in the Harvard Dataverse doi107910DVNCEMFTJ

REFERENCES

Aaronson Daniel ldquoUsing Sibling Data to Estimate the Impact of Neighborhoodson Childrens Educational Outcomesrdquo Journal of Human Resources 33 (1998)915ndash946

Altonji Joseph G and Richard K Mansfield ldquoEstimating Group Effects UsingAverages of Observables to Control for Sorting on Unobservables School andNeighborhood Effectsrdquo Yale University Working Paper 2016

Basu Sukanya ldquoAge of Entry Effects on the Education of Immigrant Children ASibling Studyrdquo Available at SSRN 1720573 2010

Bleakley Hoyt and Aimee Chin ldquoLanguage Skills and Earnings Evidence fromChildhood Immigrantsrdquo Review of Economics and Statistics 86 (2004) 481ndash496

Chetty Raj John N Friedman and Emmanuel Saez ldquoUsing Differences in Knowl-edge across Neighborhoods to Uncover the Impacts of the EITC on EarningsrdquoAmerican Economic Review 103 (2013) 2683ndash2721

Chetty Raj John N Friedman Emmanuel Saez Nicholas Turner and DannyYagan ldquoMobility Report Cards The Role of Colleges in Intergenerational Mo-bilityrdquo Stanford University Working Paper 2017

Chetty Raj and Nathaniel Hendren ldquoThe Impacts of Neighborhoods on Inter-generational Mobility II County-Level Estimatesrdquo Quarterly Journal of Eco-nomics 133 (2018a) 1163ndash1228

mdashmdashmdash ldquoReplication Code for lsquoThe Impacts of Neighborhoods on IntergenerationalMobility I Childhood Exposure Effectsrsquo and lsquoThe Impacts of Neighborhoodson Intergenerational Mobility II County-Level Estimatesrsquordquo 2018b HarvardDataverse doi107910DVNCEMFTJ

Chetty Raj Nathaniel Hendren and Lawrence F Katz ldquoThe Effects of Expo-sure to Better Neighborhoods on Children New Evidence from the Moving toOpportunity Experimentrdquo American Economic Review 106 (2016) 855ndash902

Chetty Raj Nathaniel Hendren Patrick Kline and Emmanuel Saez ldquoWhere Isthe Land of Opportunity The Geography of Intergenerational Mobility in theUnited Statesrdquo Quarterly Journal of Economics 129 (2014) 1553ndash1623

Chetty Raj Nathaniel Hendren Frina Lin Jeremy Majerovitz and BenjaminScuderi ldquoChildhood Environment and Gender Gaps in Adulthoodrdquo AmericanEconomic Review Papers and Proceedings 106 (2016) 282ndash288

Chyn Eric ldquoMoved to Opportunity The Long-Run Effect of Public Housing De-molition on Labor Market Outcomes of Childrenrdquo American Economic Review(forthcoming)

Cilke James ldquoA Profile of Non-Filersrdquo US Department of the Treasury Office ofTax Analysis Working Paper 78 1998

Clampet-Lundquist Susan and Douglas S Massey ldquoNeighborhood Effects onEconomic Self Sufficiency A Reconsideration of the Moving to OpportunityExperimentrdquo American Journal of Sociology 114 (2008) 107ndash143

Crowder Kyle and Scott J South ldquoSpatial and Temporal Dimensions of Neighbor-hood Effects on High School Graduationrdquo Social Science Research 40 (2011)87ndash106

Damm Anna Piil and Christian Dustmann ldquoDoes Growing Up in a High CrimeNeighborhood Affect Youth Criminal Behaviorrdquo American Economic Review104 (2014) 1806ndash1832

Finkelstein Amy Matthew Gentzkow and Heidi Williams ldquoSources of GeographicVariation in Health Care Evidence from Patient Migrationrdquo Quarterly Jour-nal of Economics 131 (2016) 1681ndash1726



Haider Steven and Gary Solon ldquoLife-Cycle Variation in the Association betweenCurrent and Lifetime Earningsrdquo American Economic Review 96 (2006) 1308ndash1320

Jencks Christopher and Susan E Mayer ldquoThe Social Consequences of GrowingUp in a Poor Neighborhoodrdquo National Research Council technical report 1990

Katz Lawrence F Jeffrey R Kling and Jeffrey B Liebman ldquoMoving to Opportu-nity in Boston Early Results of a Randomized Mobility Experimentrdquo Quar-terly Journal of Economics 116 (2001) 607ndash654

Kling Jeffrey R Jeffrey B Liebman and Lawrence F Katz ldquoExperimental Anal-ysis of Neighborhood Effectsrdquo Econometrica 75 (2007) 83ndash119

Ludwig Jens Greg J Duncan Lisa A Gennetian Lawrence F Katz RonaldC Kessler Jeffrey R Kling and Lisa Sanbonmatsu ldquoLong-Term Neighbor-hood Effects on Low-Income Families Evidence from Moving to OpportunityrdquoAmerican Economic Review Papers and Proceedings 103 (2013) 226ndash231

Ludwig Jens Jeffrey B Liebman Jeffrey R Kling Greg J Duncan LawrenceF Katz Ronald C Kessler and Lisa Sanbonmatsu ldquoWhat Can We Learnabout Neighborhood Effects from the Moving to Opportunity ExperimentrdquoAmerican Journal of Sociology 114 (2008) 144ndash188

Lynch John and George Davey Smith ldquoA Life Course Approach to Chronic DiseaseEpidemiologyrdquo Annual Review of Public Health 26 (2005) 1ndash35

Massey Douglas S and Nancy A Denton American Apartheid Segregation andthe Making of the Underclas (Cambridge MA Harvard University Press1993)

Oreopoulos Philip ldquoThe Long-Run Consequences of Living in a Poor Neighbor-hoodrdquo Quarterly Journal of Economics 118 (2003) 1533ndash1175

Plotnick R and S Hoffman ldquoThe Effect of Neighborhood Characteristics on YoungAdult Outcomes Alternative Estimatesrdquo Institute for Research on PovertyDiscussion Paper no 1106ndash96 1996

Rickford John R Greg J Duncan Lisa A Gennetian Ray Yun Gou RebeccaGreene Lawrence F Katz Ronald C Kessler Jeffrey R Kling Lisa San-bonmatsu Andres E Sanchez-Ordonez Matthew Sciandra Ewart Thomasand Jens Ludwig ldquoNeighborhood effects on use of African-American Vernac-ular Englishrdquo Proceedings of the National Academy of Sciences 112 (2015)11817ndash11822

Sacerdote Bruce ldquoWhen the Saints Go Marching Out Long-Term Outcomes forStudent Evacuees from Hurricanes Katrina and Ritardquo American EconomicJournal Applied Economics 4 (2012) 109ndash135

Sampson Robert J Jeffrey D Morenoff and Thomas Gannon-Rowley ldquoAssess-ing Neighborhood Effects Social Processes and New Directions in ResearchrdquoAnnual Review of Sociology 28 (2002) 443ndash478

Sharkey Patrick and Jacob W Faber ldquoWhere When Why and for Whom Do Resi-dential Contexts Matter Moving Away from the Dichotomous Understandingof Neighborhood Effectsrdquo Annual Review of Sociology 40 (2014) 559ndash579

Singleton DM and L Ryan Language Acquisition The Age Factor (ClevedonMultilingual Matters 2004)

Solon Gary ldquoIntergenerational Income Mobility in the United Statesrdquo AmericanEconomic Review 82 (1992) 393ndash408

Tolbert Charles M and Molly Sizer ldquoUS Commuting Zones and Labor MarketAreas A 1990 Updaterdquo Economic Research Service Staff Paper 9614 (1996)

van den Berg Gerard J Petter Lundborg Paul Nystedt and Dan-Olof RoothldquoCritical Periods during Childhood and Adolescencerdquo Journal of the EuropeanEconomic Association 12 (2014) 1521ndash1557

Wilson William J The Truly Disadvantaged The Inner City the Underclass andPublic Policy (Chicago University of Chicago Press 1987)

Wodtke Geoffrey T ldquoDuration and Timing of Exposure to Neighborhood Povertyand the Risk of Adolescent Parenthoodrdquo Demography 50 (2013) 1765ndash1788

Wodtke Geoffrey T David J Harding and Felix Elwert ldquoNeighborhood Effectsin Temporal Perspective The Impact of Long-Term Exposure to ConcentratedDisadvantage on High School Graduationrdquo American Sociological Review 76(2011) 713ndash736



confounding factors by comparing the outcomes of siblings within families study-ing moves triggered by displacement shocks and exploiting sharp variation inpredicted place effects across birth cohorts genders and quantiles to implementoveridentification tests The findings show that neighborhoods affect intergenera-tional mobility primarily through childhood exposure helping reconcile conflictingresults in the prior literature JEL Codes J62 C00 R00

I INTRODUCTION

To what extent are childrenrsquos economic opportunities shapedby the neighborhoods in which they grow up Despite extensiveresearch the answer to this question remains debated Observa-tional studies have documented significant variation across neigh-borhoods in economic outcomes (eg Wilson 1987 Jencks andMayer 1990 Massey and Denton 1993 Sampson Morenoff andGannon-Rowley 2002 Sharkey and Faber 2014) However exper-imental studies of families that move have traditionally foundlittle evidence that neighborhoods affect economic outcomes (egKatz Kling and Liebman 2001 Oreopoulos 2003 Ludwig et al2013)

Using deidentified tax records covering the US populationwe present new quasi-experimental evidence on the effects ofneighborhoods on intergenerational mobility that reconcile theconflicting findings of prior work and shed light on the mecha-nisms through which neighborhoods affect childrenrsquos outcomesOur analysis consists of two articles In this article we measurethe degree to which the differences in intergenerational mobilityacross areas in observational data are driven by causal effects ofplace In the second article (Chetty and Hendren 2018a) we buildon the research design developed here to construct estimates ofthe causal effect of growing up in each county in the United Stateson childrenrsquos long-term outcomes and characterize the features ofareas that produce good outcomes

Our analysis is motivated by our previous work showing thatchildrenrsquos expected incomes conditional on their parentsrsquo incomesvary substantially with the area (commuting zone or county)in which they grow up (Chetty et al 2014)1 This geographic

1 To maximize statistical precision we characterize neighborhood (or ldquoplacerdquo)effects at two broad geographies counties and commuting zones (CZs) whichare aggregations of counties that are similar to metro areas but cover the entireUnited States including rural areas Counties are much larger than the typical

















































II DATA



































































FIGURE I











FIGURE II




















bm = βm + δm





























odps in eachbin



yi = αqosm +30sum

m=9


s=1980





FIGURE III

















FIGURE IV
















yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+30sum

m=9


s=1980













yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+1987sum

s=1980



)odps




sum1988s=1980 I(si = s)α2






TA

BL

EII

CH

ILD

HO

OD

EX

PO

SU

RE

EF

FE

CT

ES

TIM

AT

ES

Dep

ende

nt

vari

able

Ch

ildrsquo

sin

com

era

nk

atag

e24

Wit

hfa

mil

yfi

xed

effe

cts

Spe

cifi

cati

on

Poo

led

Age

23

Age

lt18

No

coh

ort

Indi

vidu

alC

hil

dC

ZB

asel

ine

No

coh

ort

Tim

e-co

ntr

ols

inco

me

FE

con

trol

sva

ryin

gco

ntr

ols

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Exp

osu

reef

fect

(γ)

004

00

040

003

70

036

004

10

031

004

40

031

004

3(0

002

)(0

002

)(0

005

)(0

002

)(0

002

)(0

002

)(0

008

)(0

005

)(0

008

)

Nu

mo

fob

s1

553

021

128

777

368

732

31

553

021

155

302

11

473

218

155

302

11

553

021

155

302

1

Not

esT

his

tabl

ere

port

ses

tim

ates

ofan

nu

alch

ildh

ood

expo

sure

effe

cts

onch

ildr

enrsquos

inco

me

ran

ksat

age

24(γ

)T

he

esti

mat

esca

nbe

inte

rpre

ted

asth

eim

pact

ofsp

endi

ng

anad

diti

onal

year

ofch

ildh

ood

ina

CZ

wh

ere

chil

dren

ofpe

rman

ent

resi

den

tsh

ave

1pe

rcen

tile

poin

th

igh

erin

com

era

nks

atag

e24

Sta

nda

rder

rors

are

show

nin

pare

nth

eses

Eac

hco

lum

nre

port

ses

tim

ates

from

are

gres

sion

ofa

chil

drsquos

inco

me

ran

kat

age

24on

the

diff

eren

cebe

twee

npe

rman

ent

resi

den

tsrsquop

redi

cted

ran

ksin

the

dest

inat

ion

vers

us

the

orig

in

inte

ract

edw

ith

the

age

ofth

ech

ild

atth

eti

me

ofth

em

ove

(m)

We

perm

itse

para

teli

nea

rin

tera

ctio

ns

for

m

23an

dm

gt23

and

repo

rtth

eco

effi

cien

ton

the

inte

ract

ion

for

m

23E

ach

regr

essi

onal

soin

clu

des

addi

tion

alco

ntr

ols

spec

ified

ineq

uat

ion

(7)

Per

man

ent

resi

den

tsrsquop

redi

cted

ran

ksar

eco

nst

ruct

edu

sin

gli

nea

rre

gres

sion

sof

chil

dren

rsquosra

nks

onpa

ren

tsrsquor

anks

inea

chC

Zan

dbi

rth

coh

ort

assh

own

inF

igu

reI

Col

um

n(1

)re

port

sth

ees

tim

ate

ofγ

from

equ

atio

n(7

)u

sin

gal

lch

ildr

enin

the

prim

ary

anal

ysis

sam

ple

ofon

e-ti

me

mov

ers

defi

ned

inth

en

otes

toT

able

I(P

anel

C)

Col

um

ns

(2)

and

(3)

rest

rict

the

sam

ple

toth

ose

wh

om

ove

ator

befo

reag

e23

or18

In

colu

mn

(4)

we

excl

ude

the

coh

ort

inte

ract

ion

sw

ith

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

loca

tion

and

inst

ead

incl

ude

asi

ngl

eco

ntr

olfo

rth

epr

edic

ted

outc

omes

ofpe

rman

ent

resi

den

tsin

the

orig

inC

olu

mn

(5)r

epli

cate

sco

lum

n(1

)u

sin

gin

divi

dual

inco

me

ran

ks(r

ath

erth

anh

ouse

hol

din

com

era

nks

)to

mea

sure

both

the

chil

drsquos

outc

ome

and

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

Col

um

n(6

)ad

dsfi

xed

effe

cts

for

the

chil

drsquos

CZ

in20

12to

the

spec

ifica

tion

inco

lum

n(1

)n

ote

that

this

spec

ifica

tion

has

asl

igh

tly

smal

ler

nu

mbe

rof

obse

rvat

ion

sbe

cau

seZ

IPco

dein

form

atio

nis

mis

sin

gfo

rso

me

chil

dren

in20

12(e

g

ach

ild

wit

hn

oea

rnin

gsor

taxa

ble

inco

me)

Col

um

n(7

)ad

dsfa

mil

yfi

xed

effe

cts

toth

eba

seli

ne

spec

ifica

tion

inco

lum

n(1

)C

olu

mn

(8)

adds

fam

ily

fixe

def

fect

sto

the

spec

ifica

tion

inco

lum

n(4

)th

atdo

esn

otin

clu

deco

hor

t-va

ryin

gin

terc

epts

C

olu

mn

(9)

adds

con

trol

sfo

rch

ange

sin

pare

nta

lm

arit

alst

atu

san

din

com

era

nk

inth

eye

arbe

fore

vers

us

afte

rth

em

ove

alon

gw

ith

thei

rin

tera

ctio

ns

wit

hth

eag

eof

the

chil

dat

the

tim

eof

the

mov

ean

din

dica

tors

for

mov

ing

abov

ean

dbe

low

age

23t

oth

esp

ecifi

cati

onin

colu

mn

(7)





IVC Mechanisms
















IVD Summary




















FIGURE V






















Kz34








FIGURE VI

























FIGURE VII


















(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES

























































































II DATA



































































FIGURE I











FIGURE II




















bm = βm + δm





























odps in eachbin



yi = αqosm +30sum

m=9


s=1980





FIGURE III

















FIGURE IV
















yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+30sum

m=9


s=1980













yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+1987sum

s=1980



)odps




sum1988s=1980 I(si = s)α2






TA

BL

EII

CH

ILD

HO

OD

EX

PO

SU

RE

EF

FE

CT

ES

TIM

AT

ES

Dep

ende

nt

vari

able

Ch

ildrsquo

sin

com

era

nk

atag

e24

Wit

hfa

mil

yfi

xed

effe

cts

Spe

cifi

cati

on

Poo

led

Age

23

Age

lt18

No

coh

ort

Indi

vidu

alC

hil

dC

ZB

asel

ine

No

coh

ort

Tim

e-co

ntr

ols

inco

me

FE

con

trol

sva

ryin

gco

ntr

ols

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Exp

osu

reef

fect

(γ)

004

00

040

003

70

036

004

10

031

004

40

031

004

3(0

002

)(0

002

)(0

005

)(0

002

)(0

002

)(0

002

)(0

008

)(0

005

)(0

008

)

Nu

mo

fob

s1

553

021

128

777

368

732

31

553

021

155

302

11

473

218

155

302

11

553

021

155

302

1

Not

esT

his

tabl

ere

port

ses

tim

ates

ofan

nu

alch

ildh

ood

expo

sure

effe

cts

onch

ildr

enrsquos

inco

me

ran

ksat

age

24(γ

)T

he

esti

mat

esca

nbe

inte

rpre

ted

asth

eim

pact

ofsp

endi

ng

anad

diti

onal

year

ofch

ildh

ood

ina

CZ

wh

ere

chil

dren

ofpe

rman

ent

resi

den

tsh

ave

1pe

rcen

tile

poin

th

igh

erin

com

era

nks

atag

e24

Sta

nda

rder

rors

are

show

nin

pare

nth

eses

Eac

hco

lum

nre

port

ses

tim

ates

from

are

gres

sion

ofa

chil

drsquos

inco

me

ran

kat

age

24on

the

diff

eren

cebe

twee

npe

rman

ent

resi

den

tsrsquop

redi

cted

ran

ksin

the

dest

inat

ion

vers

us

the

orig

in

inte

ract

edw

ith

the

age

ofth

ech

ild

atth

eti

me

ofth

em

ove

(m)

We

perm

itse

para

teli

nea

rin

tera

ctio

ns

for

m

23an

dm

gt23

and

repo

rtth

eco

effi

cien

ton

the

inte

ract

ion

for

m

23E

ach

regr

essi

onal

soin

clu

des

addi

tion

alco

ntr

ols

spec

ified

ineq

uat

ion

(7)

Per

man

ent

resi

den

tsrsquop

redi

cted

ran

ksar

eco

nst

ruct

edu

sin

gli

nea

rre

gres

sion

sof

chil

dren

rsquosra

nks

onpa

ren

tsrsquor

anks

inea

chC

Zan

dbi

rth

coh

ort

assh

own

inF

igu

reI

Col

um

n(1

)re

port

sth

ees

tim

ate

ofγ

from

equ

atio

n(7

)u

sin

gal

lch

ildr

enin

the

prim

ary

anal

ysis

sam

ple

ofon

e-ti

me

mov

ers

defi

ned

inth

en

otes

toT

able

I(P

anel

C)

Col

um

ns

(2)

and

(3)

rest

rict

the

sam

ple

toth

ose

wh

om

ove

ator

befo

reag

e23

or18

In

colu

mn

(4)

we

excl

ude

the

coh

ort

inte

ract

ion

sw

ith

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

loca

tion

and

inst

ead

incl

ude

asi

ngl

eco

ntr

olfo

rth

epr

edic

ted

outc

omes

ofpe

rman

ent

resi

den

tsin

the

orig

inC

olu

mn

(5)r

epli

cate

sco

lum

n(1

)u

sin

gin

divi

dual

inco

me

ran

ks(r

ath

erth

anh

ouse

hol

din

com

era

nks

)to

mea

sure

both

the

chil

drsquos

outc

ome

and

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

Col

um

n(6

)ad

dsfi

xed

effe

cts

for

the

chil

drsquos

CZ

in20

12to

the

spec

ifica

tion

inco

lum

n(1

)n

ote

that

this

spec

ifica

tion

has

asl

igh

tly

smal

ler

nu

mbe

rof

obse

rvat

ion

sbe

cau

seZ

IPco

dein

form

atio

nis

mis

sin

gfo

rso

me

chil

dren

in20

12(e

g

ach

ild

wit

hn

oea

rnin

gsor

taxa

ble

inco

me)

Col

um

n(7

)ad

dsfa

mil

yfi

xed

effe

cts

toth

eba

seli

ne

spec

ifica

tion

inco

lum

n(1

)C

olu

mn

(8)

adds

fam

ily

fixe

def

fect

sto

the

spec

ifica

tion

inco

lum

n(4

)th

atdo

esn

otin

clu

deco

hor

t-va

ryin

gin

terc

epts

C

olu

mn

(9)

adds

con

trol

sfo

rch

ange

sin

pare

nta

lm

arit

alst

atu

san

din

com

era

nk

inth

eye

arbe

fore

vers

us

afte

rth

em

ove

alon

gw

ith

thei

rin

tera

ctio

ns

wit

hth

eag

eof

the

chil

dat

the

tim

eof

the

mov

ean

din

dica

tors

for

mov

ing

abov

ean

dbe

low

age

23t

oth

esp

ecifi

cati

onin

colu

mn

(7)





IVC Mechanisms
















IVD Summary




















FIGURE V






















Kz34








FIGURE VI

























FIGURE VII


















(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES






































































































FIGURE I











FIGURE II




















bm = βm + δm





























odps in eachbin



yi = αqosm +30sum

m=9


s=1980





FIGURE III

















FIGURE IV
















yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+30sum

m=9


s=1980













yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+1987sum

s=1980



)odps




sum1988s=1980 I(si = s)α2






TA

BL

EII

CH

ILD

HO

OD

EX

PO

SU

RE

EF

FE

CT

ES

TIM

AT

ES

Dep

ende

nt

vari

able

Ch

ildrsquo

sin

com

era

nk

atag

e24

Wit

hfa

mil

yfi

xed

effe

cts

Spe

cifi

cati

on

Poo

led

Age

23

Age

lt18

No

coh

ort

Indi

vidu

alC

hil

dC

ZB

asel

ine

No

coh

ort

Tim

e-co

ntr

ols

inco

me

FE

con

trol

sva

ryin

gco

ntr

ols

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Exp

osu

reef

fect

(γ)

004

00

040

003

70

036

004

10

031

004

40

031

004

3(0

002

)(0

002

)(0

005

)(0

002

)(0

002

)(0

002

)(0

008

)(0

005

)(0

008

)

Nu

mo

fob

s1

553

021

128

777

368

732

31

553

021

155

302

11

473

218

155

302

11

553

021

155

302

1

Not

esT

his

tabl

ere

port

ses

tim

ates

ofan

nu

alch

ildh

ood

expo

sure

effe

cts

onch

ildr

enrsquos

inco

me

ran

ksat

age

24(γ

)T

he

esti

mat

esca

nbe

inte

rpre

ted

asth

eim

pact

ofsp

endi

ng

anad

diti

onal

year

ofch

ildh

ood

ina

CZ

wh

ere

chil

dren

ofpe

rman

ent

resi

den

tsh

ave

1pe

rcen

tile

poin

th

igh

erin

com

era

nks

atag

e24

Sta

nda

rder

rors

are

show

nin

pare

nth

eses

Eac

hco

lum

nre

port

ses

tim

ates

from

are

gres

sion

ofa

chil

drsquos

inco

me

ran

kat

age

24on

the

diff

eren

cebe

twee

npe

rman

ent

resi

den

tsrsquop

redi

cted

ran

ksin

the

dest

inat

ion

vers

us

the

orig

in

inte

ract

edw

ith

the

age

ofth

ech

ild

atth

eti

me

ofth

em

ove

(m)

We

perm

itse

para

teli

nea

rin

tera

ctio

ns

for

m

23an

dm

gt23

and

repo

rtth

eco

effi

cien

ton

the

inte

ract

ion

for

m

23E

ach

regr

essi

onal

soin

clu

des

addi

tion

alco

ntr

ols

spec

ified

ineq

uat

ion

(7)

Per

man

ent

resi

den

tsrsquop

redi

cted

ran

ksar

eco

nst

ruct

edu

sin

gli

nea

rre

gres

sion

sof

chil

dren

rsquosra

nks

onpa

ren

tsrsquor

anks

inea

chC

Zan

dbi

rth

coh

ort

assh

own

inF

igu

reI

Col

um

n(1

)re

port

sth

ees

tim

ate

ofγ

from

equ

atio

n(7

)u

sin

gal

lch

ildr

enin

the

prim

ary

anal

ysis

sam

ple

ofon

e-ti

me

mov

ers

defi

ned

inth

en

otes

toT

able

I(P

anel

C)

Col

um

ns

(2)

and

(3)

rest

rict

the

sam

ple

toth

ose

wh

om

ove

ator

befo

reag

e23

or18

In

colu

mn

(4)

we

excl

ude

the

coh

ort

inte

ract

ion

sw

ith

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

loca

tion

and

inst

ead

incl

ude

asi

ngl

eco

ntr

olfo

rth

epr

edic

ted

outc

omes

ofpe

rman

ent

resi

den

tsin

the

orig

inC

olu

mn

(5)r

epli

cate

sco

lum

n(1

)u

sin

gin

divi

dual

inco

me

ran

ks(r

ath

erth

anh

ouse

hol

din

com

era

nks

)to

mea

sure

both

the

chil

drsquos

outc

ome

and

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

Col

um

n(6

)ad

dsfi

xed

effe

cts

for

the

chil

drsquos

CZ

in20

12to

the

spec

ifica

tion

inco

lum

n(1

)n

ote

that

this

spec

ifica

tion

has

asl

igh

tly

smal

ler

nu

mbe

rof

obse

rvat

ion

sbe

cau

seZ

IPco

dein

form

atio

nis

mis

sin

gfo

rso

me

chil

dren

in20

12(e

g

ach

ild

wit

hn

oea

rnin

gsor

taxa

ble

inco

me)

Col

um

n(7

)ad

dsfa

mil

yfi

xed

effe

cts

toth

eba

seli

ne

spec

ifica

tion

inco

lum

n(1

)C

olu

mn

(8)

adds

fam

ily

fixe

def

fect

sto

the

spec

ifica

tion

inco

lum

n(4

)th

atdo

esn

otin

clu

deco

hor

t-va

ryin

gin

terc

epts

C

olu

mn

(9)

adds

con

trol

sfo

rch

ange

sin

pare

nta

lm

arit

alst

atu

san

din

com

era

nk

inth

eye

arbe

fore

vers

us

afte

rth

em

ove

alon

gw

ith

thei

rin

tera

ctio

ns

wit

hth

eag

eof

the

chil

dat

the

tim

eof

the

mov

ean

din

dica

tors

for

mov

ing

abov

ean

dbe

low

age

23t

oth

esp

ecifi

cati

onin

colu

mn

(7)





IVC Mechanisms
















IVD Summary




















FIGURE V






















Kz34








FIGURE VI

























FIGURE VII


















(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES











































FIGURE II




















bm = βm + δm





























odps in eachbin



yi = αqosm +30sum

m=9


s=1980





FIGURE III

















FIGURE IV
















yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+30sum

m=9


s=1980













yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+1987sum

s=1980



)odps




sum1988s=1980 I(si = s)α2






TA

BL

EII

CH

ILD

HO

OD

EX

PO

SU

RE

EF

FE

CT

ES

TIM

AT

ES

Dep

ende

nt

vari

able

Ch

ildrsquo

sin

com

era

nk

atag

e24

Wit

hfa

mil

yfi

xed

effe

cts

Spe

cifi

cati

on

Poo

led

Age

23

Age

lt18

No

coh

ort

Indi

vidu

alC

hil

dC

ZB

asel

ine

No

coh

ort

Tim

e-co

ntr

ols

inco

me

FE

con

trol

sva

ryin

gco

ntr

ols

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Exp

osu

reef

fect

(γ)

004

00

040

003

70

036

004

10

031

004

40

031

004

3(0

002

)(0

002

)(0

005

)(0

002

)(0

002

)(0

002

)(0

008

)(0

005

)(0

008

)

Nu

mo

fob

s1

553

021

128

777

368

732

31

553

021

155

302

11

473

218

155

302

11

553

021

155

302

1

Not

esT

his

tabl

ere

port

ses

tim

ates

ofan

nu

alch

ildh

ood

expo

sure

effe

cts

onch

ildr

enrsquos

inco

me

ran

ksat

age

24(γ

)T

he

esti

mat

esca

nbe

inte

rpre

ted

asth

eim

pact

ofsp

endi

ng

anad

diti

onal

year

ofch

ildh

ood

ina

CZ

wh

ere

chil

dren

ofpe

rman

ent

resi

den

tsh

ave

1pe

rcen

tile

poin

th

igh

erin

com

era

nks

atag

e24

Sta

nda

rder

rors

are

show

nin

pare

nth

eses

Eac

hco

lum

nre

port

ses

tim

ates

from

are

gres

sion

ofa

chil

drsquos

inco

me

ran

kat

age

24on

the

diff

eren

cebe

twee

npe

rman

ent

resi

den

tsrsquop

redi

cted

ran

ksin

the

dest

inat

ion

vers

us

the

orig

in

inte

ract

edw

ith

the

age

ofth

ech

ild

atth

eti

me

ofth

em

ove

(m)

We

perm

itse

para

teli

nea

rin

tera

ctio

ns

for

m

23an

dm

gt23

and

repo

rtth

eco

effi

cien

ton

the

inte

ract

ion

for

m

23E

ach

regr

essi

onal

soin

clu

des

addi

tion

alco

ntr

ols

spec

ified

ineq

uat

ion

(7)

Per

man

ent

resi

den

tsrsquop

redi

cted

ran

ksar

eco

nst

ruct

edu

sin

gli

nea

rre

gres

sion

sof

chil

dren

rsquosra

nks

onpa

ren

tsrsquor

anks

inea

chC

Zan

dbi

rth

coh

ort

assh

own

inF

igu

reI

Col

um

n(1

)re

port

sth

ees

tim

ate

ofγ

from

equ

atio

n(7

)u

sin

gal

lch

ildr

enin

the

prim

ary

anal

ysis

sam

ple

ofon

e-ti

me

mov

ers

defi

ned

inth

en

otes

toT

able

I(P

anel

C)

Col

um

ns

(2)

and

(3)

rest

rict

the

sam

ple

toth

ose

wh

om

ove

ator

befo

reag

e23

or18

In

colu

mn

(4)

we

excl

ude

the

coh

ort

inte

ract

ion

sw

ith

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

loca

tion

and

inst

ead

incl

ude

asi

ngl

eco

ntr

olfo

rth

epr

edic

ted

outc

omes

ofpe

rman

ent

resi

den

tsin

the

orig

inC

olu

mn

(5)r

epli

cate

sco

lum

n(1

)u

sin

gin

divi

dual

inco

me

ran

ks(r

ath

erth

anh

ouse

hol

din

com

era

nks

)to

mea

sure

both

the

chil

drsquos

outc

ome

and

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

Col

um

n(6

)ad

dsfi

xed

effe

cts

for

the

chil

drsquos

CZ

in20

12to

the

spec

ifica

tion

inco

lum

n(1

)n

ote

that

this

spec

ifica

tion

has

asl

igh

tly

smal

ler

nu

mbe

rof

obse

rvat

ion

sbe

cau

seZ

IPco

dein

form

atio

nis

mis

sin

gfo

rso

me

chil

dren

in20

12(e

g

ach

ild

wit

hn

oea

rnin

gsor

taxa

ble

inco

me)

Col

um

n(7

)ad

dsfa

mil

yfi

xed

effe

cts

toth

eba

seli

ne

spec

ifica

tion

inco

lum

n(1

)C

olu

mn

(8)

adds

fam

ily

fixe

def

fect

sto

the

spec

ifica

tion

inco

lum

n(4

)th

atdo

esn

otin

clu

deco

hor

t-va

ryin

gin

terc

epts

C

olu

mn

(9)

adds

con

trol

sfo

rch

ange

sin

pare

nta

lm

arit

alst

atu

san

din

com

era

nk

inth

eye

arbe

fore

vers

us

afte

rth

em

ove

alon

gw

ith

thei

rin

tera

ctio

ns

wit

hth

eag

eof

the

chil

dat

the

tim

eof

the

mov

ean

din

dica

tors

for

mov

ing

abov

ean

dbe

low

age

23t

oth

esp

ecifi

cati

onin

colu

mn

(7)





IVC Mechanisms
















IVD Summary




















FIGURE V






















Kz34








FIGURE VI

























FIGURE VII


















(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES


























































bm = βm + δm





























odps in eachbin



yi = αqosm +30sum

m=9


s=1980





FIGURE III

















FIGURE IV
















yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+30sum

m=9


s=1980













yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+1987sum

s=1980



)odps




sum1988s=1980 I(si = s)α2






TA

BL

EII

CH

ILD

HO

OD

EX

PO

SU

RE

EF

FE

CT

ES

TIM

AT

ES

Dep

ende

nt

vari

able

Ch

ildrsquo

sin

com

era

nk

atag

e24

Wit

hfa

mil

yfi

xed

effe

cts

Spe

cifi

cati

on

Poo

led

Age

23

Age

lt18

No

coh

ort

Indi

vidu

alC

hil

dC

ZB

asel

ine

No

coh

ort

Tim

e-co

ntr

ols

inco

me

FE

con

trol

sva

ryin

gco

ntr

ols

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Exp

osu

reef

fect

(γ)

004

00

040

003

70

036

004

10

031

004

40

031

004

3(0

002

)(0

002

)(0

005

)(0

002

)(0

002

)(0

002

)(0

008

)(0

005

)(0

008

)

Nu

mo

fob

s1

553

021

128

777

368

732

31

553

021

155

302

11

473

218

155

302

11

553

021

155

302

1

Not

esT

his

tabl

ere

port

ses

tim

ates

ofan

nu

alch

ildh

ood

expo

sure

effe

cts

onch

ildr

enrsquos

inco

me

ran

ksat

age

24(γ

)T

he

esti

mat

esca

nbe

inte

rpre

ted

asth

eim

pact

ofsp

endi

ng

anad

diti

onal

year

ofch

ildh

ood

ina

CZ

wh

ere

chil

dren

ofpe

rman

ent

resi

den

tsh

ave

1pe

rcen

tile

poin

th

igh

erin

com

era

nks

atag

e24

Sta

nda

rder

rors

are

show

nin

pare

nth

eses

Eac

hco

lum

nre

port

ses

tim

ates

from

are

gres

sion

ofa

chil

drsquos

inco

me

ran

kat

age

24on

the

diff

eren

cebe

twee

npe

rman

ent

resi

den

tsrsquop

redi

cted

ran

ksin

the

dest

inat

ion

vers

us

the

orig

in

inte

ract

edw

ith

the

age

ofth

ech

ild

atth

eti

me

ofth

em

ove

(m)

We

perm

itse

para

teli

nea

rin

tera

ctio

ns

for

m

23an

dm

gt23

and

repo

rtth

eco

effi

cien

ton

the

inte

ract

ion

for

m

23E

ach

regr

essi

onal

soin

clu

des

addi

tion

alco

ntr

ols

spec

ified

ineq

uat

ion

(7)

Per

man

ent

resi

den

tsrsquop

redi

cted

ran

ksar

eco

nst

ruct

edu

sin

gli

nea

rre

gres

sion

sof

chil

dren

rsquosra

nks

onpa

ren

tsrsquor

anks

inea

chC

Zan

dbi

rth

coh

ort

assh

own

inF

igu

reI

Col

um

n(1

)re

port

sth

ees

tim

ate

ofγ

from

equ

atio

n(7

)u

sin

gal

lch

ildr

enin

the

prim

ary

anal

ysis

sam

ple

ofon

e-ti

me

mov

ers

defi

ned

inth

en

otes

toT

able

I(P

anel

C)

Col

um

ns

(2)

and

(3)

rest

rict

the

sam

ple

toth

ose

wh

om

ove

ator

befo

reag

e23

or18

In

colu

mn

(4)

we

excl

ude

the

coh

ort

inte

ract

ion

sw

ith

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

loca

tion

and

inst

ead

incl

ude

asi

ngl

eco

ntr

olfo

rth

epr

edic

ted

outc

omes

ofpe

rman

ent

resi

den

tsin

the

orig

inC

olu

mn

(5)r

epli

cate

sco

lum

n(1

)u

sin

gin

divi

dual

inco

me

ran

ks(r

ath

erth

anh

ouse

hol

din

com

era

nks

)to

mea

sure

both

the

chil

drsquos

outc

ome

and

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

Col

um

n(6

)ad

dsfi

xed

effe

cts

for

the

chil

drsquos

CZ

in20

12to

the

spec

ifica

tion

inco

lum

n(1

)n

ote

that

this

spec

ifica

tion

has

asl

igh

tly

smal

ler

nu

mbe

rof

obse

rvat

ion

sbe

cau

seZ

IPco

dein

form

atio

nis

mis

sin

gfo

rso

me

chil

dren

in20

12(e

g

ach

ild

wit

hn

oea

rnin

gsor

taxa

ble

inco

me)

Col

um

n(7

)ad

dsfa

mil

yfi

xed

effe

cts

toth

eba

seli

ne

spec

ifica

tion

inco

lum

n(1

)C

olu

mn

(8)

adds

fam

ily

fixe

def

fect

sto

the

spec

ifica

tion

inco

lum

n(4

)th

atdo

esn

otin

clu

deco

hor

t-va

ryin

gin

terc

epts

C

olu

mn

(9)

adds

con

trol

sfo

rch

ange

sin

pare

nta

lm

arit

alst

atu

san

din

com

era

nk

inth

eye

arbe

fore

vers

us

afte

rth

em

ove

alon

gw

ith

thei

rin

tera

ctio

ns

wit

hth

eag

eof

the

chil

dat

the

tim

eof

the

mov

ean

din

dica

tors

for

mov

ing

abov

ean

dbe

low

age

23t

oth

esp

ecifi

cati

onin

colu

mn

(7)





IVC Mechanisms
















IVD Summary




















FIGURE V






















Kz34








FIGURE VI

























FIGURE VII


















(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES



































































odps in eachbin



yi = αqosm +30sum

m=9


s=1980





FIGURE III

















FIGURE IV
















yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+30sum

m=9


s=1980













yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+1987sum

s=1980



)odps




sum1988s=1980 I(si = s)α2






TA

BL

EII

CH

ILD

HO

OD

EX

PO

SU

RE

EF

FE

CT

ES

TIM

AT

ES

Dep

ende

nt

vari

able

Ch

ildrsquo

sin

com

era

nk

atag

e24

Wit

hfa

mil

yfi

xed

effe

cts

Spe

cifi

cati

on

Poo

led

Age

23

Age

lt18

No

coh

ort

Indi

vidu

alC

hil

dC

ZB

asel

ine

No

coh

ort

Tim

e-co

ntr

ols

inco

me

FE

con

trol

sva

ryin

gco

ntr

ols

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Exp

osu

reef

fect

(γ)

004

00

040

003

70

036

004

10

031

004

40

031

004

3(0

002

)(0

002

)(0

005

)(0

002

)(0

002

)(0

002

)(0

008

)(0

005

)(0

008

)

Nu

mo

fob

s1

553

021

128

777

368

732

31

553

021

155

302

11

473

218

155

302

11

553

021

155

302

1

Not

esT

his

tabl

ere

port

ses

tim

ates

ofan

nu

alch

ildh

ood

expo

sure

effe

cts

onch

ildr

enrsquos

inco

me

ran

ksat

age

24(γ

)T

he

esti

mat

esca

nbe

inte

rpre

ted

asth

eim

pact

ofsp

endi

ng

anad

diti

onal

year

ofch

ildh

ood

ina

CZ

wh

ere

chil

dren

ofpe

rman

ent

resi

den

tsh

ave

1pe

rcen

tile

poin

th

igh

erin

com

era

nks

atag

e24

Sta

nda

rder

rors

are

show

nin

pare

nth

eses

Eac

hco

lum

nre

port

ses

tim

ates

from

are

gres

sion

ofa

chil

drsquos

inco

me

ran

kat

age

24on

the

diff

eren

cebe

twee

npe

rman

ent

resi

den

tsrsquop

redi

cted

ran

ksin

the

dest

inat

ion

vers

us

the

orig

in

inte

ract

edw

ith

the

age

ofth

ech

ild

atth

eti

me

ofth

em

ove

(m)

We

perm

itse

para

teli

nea

rin

tera

ctio

ns

for

m

23an

dm

gt23

and

repo

rtth

eco

effi

cien

ton

the

inte

ract

ion

for

m

23E

ach

regr

essi

onal

soin

clu

des

addi

tion

alco

ntr

ols

spec

ified

ineq

uat

ion

(7)

Per

man

ent

resi

den

tsrsquop

redi

cted

ran

ksar

eco

nst

ruct

edu

sin

gli

nea

rre

gres

sion

sof

chil

dren

rsquosra

nks

onpa

ren

tsrsquor

anks

inea

chC

Zan

dbi

rth

coh

ort

assh

own

inF

igu

reI

Col

um

n(1

)re

port

sth

ees

tim

ate

ofγ

from

equ

atio

n(7

)u

sin

gal

lch

ildr

enin

the

prim

ary

anal

ysis

sam

ple

ofon

e-ti

me

mov

ers

defi

ned

inth

en

otes

toT

able

I(P

anel

C)

Col

um

ns

(2)

and

(3)

rest

rict

the

sam

ple

toth

ose

wh

om

ove

ator

befo

reag

e23

or18

In

colu

mn

(4)

we

excl

ude

the

coh

ort

inte

ract

ion

sw

ith

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

loca

tion

and

inst

ead

incl

ude

asi

ngl

eco

ntr

olfo

rth

epr

edic

ted

outc

omes

ofpe

rman

ent

resi

den

tsin

the

orig

inC

olu

mn

(5)r

epli

cate

sco

lum

n(1

)u

sin

gin

divi

dual

inco

me

ran

ks(r

ath

erth

anh

ouse

hol

din

com

era

nks

)to

mea

sure

both

the

chil

drsquos

outc

ome

and

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

Col

um

n(6

)ad

dsfi

xed

effe

cts

for

the

chil

drsquos

CZ

in20

12to

the

spec

ifica

tion

inco

lum

n(1

)n

ote

that

this

spec

ifica

tion

has

asl

igh

tly

smal

ler

nu

mbe

rof

obse

rvat

ion

sbe

cau

seZ

IPco

dein

form

atio

nis

mis

sin

gfo

rso

me

chil

dren

in20

12(e

g

ach

ild

wit

hn

oea

rnin

gsor

taxa

ble

inco

me)

Col

um

n(7

)ad

dsfa

mil

yfi

xed

effe

cts

toth

eba

seli

ne

spec

ifica

tion

inco

lum

n(1

)C

olu

mn

(8)

adds

fam

ily

fixe

def

fect

sto

the

spec

ifica

tion

inco

lum

n(4

)th

atdo

esn

otin

clu

deco

hor

t-va

ryin

gin

terc

epts

C

olu

mn

(9)

adds

con

trol

sfo

rch

ange

sin

pare

nta

lm

arit

alst

atu

san

din

com

era

nk

inth

eye

arbe

fore

vers

us

afte

rth

em

ove

alon

gw

ith

thei

rin

tera

ctio

ns

wit

hth

eag

eof

the

chil

dat

the

tim

eof

the

mov

ean

din

dica

tors

for

mov

ing

abov

ean

dbe

low

age

23t

oth

esp

ecifi

cati

onin

colu

mn

(7)





IVC Mechanisms
















IVD Summary




















FIGURE V






















Kz34








FIGURE VI

























FIGURE VII


















(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES











































FIGURE III

















FIGURE IV
















yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+30sum

m=9


s=1980













yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+1987sum

s=1980



)odps




sum1988s=1980 I(si = s)α2






TA

BL

EII

CH

ILD

HO

OD

EX

PO

SU

RE

EF

FE

CT

ES

TIM

AT

ES

Dep

ende

nt

vari

able

Ch

ildrsquo

sin

com

era

nk

atag

e24

Wit

hfa

mil

yfi

xed

effe

cts

Spe

cifi

cati

on

Poo

led

Age

23

Age

lt18

No

coh

ort

Indi

vidu

alC

hil

dC

ZB

asel

ine

No

coh

ort

Tim

e-co

ntr

ols

inco

me

FE

con

trol

sva

ryin

gco

ntr

ols

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Exp

osu

reef

fect

(γ)

004

00

040

003

70

036

004

10

031

004

40

031

004

3(0

002

)(0

002

)(0

005

)(0

002

)(0

002

)(0

002

)(0

008

)(0

005

)(0

008

)

Nu

mo

fob

s1

553

021

128

777

368

732

31

553

021

155

302

11

473

218

155

302

11

553

021

155

302

1

Not

esT

his

tabl

ere

port

ses

tim

ates

ofan

nu

alch

ildh

ood

expo

sure

effe

cts

onch

ildr

enrsquos

inco

me

ran

ksat

age

24(γ

)T

he

esti

mat

esca

nbe

inte

rpre

ted

asth

eim

pact

ofsp

endi

ng

anad

diti

onal

year

ofch

ildh

ood

ina

CZ

wh

ere

chil

dren

ofpe

rman

ent

resi

den

tsh

ave

1pe

rcen

tile

poin

th

igh

erin

com

era

nks

atag

e24

Sta

nda

rder

rors

are

show

nin

pare

nth

eses

Eac

hco

lum

nre

port

ses

tim

ates

from

are

gres

sion

ofa

chil

drsquos

inco

me

ran

kat

age

24on

the

diff

eren

cebe

twee

npe

rman

ent

resi

den

tsrsquop

redi

cted

ran

ksin

the

dest

inat

ion

vers

us

the

orig

in

inte

ract

edw

ith

the

age

ofth

ech

ild

atth

eti

me

ofth

em

ove

(m)

We

perm

itse

para

teli

nea

rin

tera

ctio

ns

for

m

23an

dm

gt23

and

repo

rtth

eco

effi

cien

ton

the

inte

ract

ion

for

m

23E

ach

regr

essi

onal

soin

clu

des

addi

tion

alco

ntr

ols

spec

ified

ineq

uat

ion

(7)

Per

man

ent

resi

den

tsrsquop

redi

cted

ran

ksar

eco

nst

ruct

edu

sin

gli

nea

rre

gres

sion

sof

chil

dren

rsquosra

nks

onpa

ren

tsrsquor

anks

inea

chC

Zan

dbi

rth

coh

ort

assh

own

inF

igu

reI

Col

um

n(1

)re

port

sth

ees

tim

ate

ofγ

from

equ

atio

n(7

)u

sin

gal

lch

ildr

enin

the

prim

ary

anal

ysis

sam

ple

ofon

e-ti

me

mov

ers

defi

ned

inth

en

otes

toT

able

I(P

anel

C)

Col

um

ns

(2)

and

(3)

rest

rict

the

sam

ple

toth

ose

wh

om

ove

ator

befo

reag

e23

or18

In

colu

mn

(4)

we

excl

ude

the

coh

ort

inte

ract

ion

sw

ith

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

loca

tion

and

inst

ead

incl

ude

asi

ngl

eco

ntr

olfo

rth

epr

edic

ted

outc

omes

ofpe

rman

ent

resi

den

tsin

the

orig

inC

olu

mn

(5)r

epli

cate

sco

lum

n(1

)u

sin

gin

divi

dual

inco

me

ran

ks(r

ath

erth

anh

ouse

hol

din

com

era

nks

)to

mea

sure

both

the

chil

drsquos

outc

ome

and

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

Col

um

n(6

)ad

dsfi

xed

effe

cts

for

the

chil

drsquos

CZ

in20

12to

the

spec

ifica

tion

inco

lum

n(1

)n

ote

that

this

spec

ifica

tion

has

asl

igh

tly

smal

ler

nu

mbe

rof

obse

rvat

ion

sbe

cau

seZ

IPco

dein

form

atio

nis

mis

sin

gfo

rso

me

chil

dren

in20

12(e

g

ach

ild

wit

hn

oea

rnin

gsor

taxa

ble

inco

me)

Col

um

n(7

)ad

dsfa

mil

yfi

xed

effe

cts

toth

eba

seli

ne

spec

ifica

tion

inco

lum

n(1

)C

olu

mn

(8)

adds

fam

ily

fixe

def

fect

sto

the

spec

ifica

tion

inco

lum

n(4

)th

atdo

esn

otin

clu

deco

hor

t-va

ryin

gin

terc

epts

C

olu

mn

(9)

adds

con

trol

sfo

rch

ange

sin

pare

nta

lm

arit

alst

atu

san

din

com

era

nk

inth

eye

arbe

fore

vers

us

afte

rth

em

ove

alon

gw

ith

thei

rin

tera

ctio

ns

wit

hth

eag

eof

the

chil

dat

the

tim

eof

the

mov

ean

din

dica

tors

for

mov

ing

abov

ean

dbe

low

age

23t

oth

esp

ecifi

cati

onin

colu

mn

(7)





IVC Mechanisms
















IVD Summary




















FIGURE V






















Kz34








FIGURE VI

























FIGURE VII


















(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES


















































FIGURE IV
















yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+30sum

m=9


s=1980













yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+1987sum

s=1980



)odps




sum1988s=1980 I(si = s)α2






TA

BL

EII

CH

ILD

HO

OD

EX

PO

SU

RE

EF

FE

CT

ES

TIM

AT

ES

Dep

ende

nt

vari

able

Ch

ildrsquo

sin

com

era

nk

atag

e24

Wit

hfa

mil

yfi

xed

effe

cts

Spe

cifi

cati

on

Poo

led

Age

23

Age

lt18

No

coh

ort

Indi

vidu

alC

hil

dC

ZB

asel

ine

No

coh

ort

Tim

e-co

ntr

ols

inco

me

FE

con

trol

sva

ryin

gco

ntr

ols

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Exp

osu

reef

fect

(γ)

004

00

040

003

70

036

004

10

031

004

40

031

004

3(0

002

)(0

002

)(0

005

)(0

002

)(0

002

)(0

002

)(0

008

)(0

005

)(0

008

)

Nu

mo

fob

s1

553

021

128

777

368

732

31

553

021

155

302

11

473

218

155

302

11

553

021

155

302

1

Not

esT

his

tabl

ere

port

ses

tim

ates

ofan

nu

alch

ildh

ood

expo

sure

effe

cts

onch

ildr

enrsquos

inco

me

ran

ksat

age

24(γ

)T

he

esti

mat

esca

nbe

inte

rpre

ted

asth

eim

pact

ofsp

endi

ng

anad

diti

onal

year

ofch

ildh

ood

ina

CZ

wh

ere

chil

dren

ofpe

rman

ent

resi

den

tsh

ave

1pe

rcen

tile

poin

th

igh

erin

com

era

nks

atag

e24

Sta

nda

rder

rors

are

show

nin

pare

nth

eses

Eac

hco

lum

nre

port

ses

tim

ates

from

are

gres

sion

ofa

chil

drsquos

inco

me

ran

kat

age

24on

the

diff

eren

cebe

twee

npe

rman

ent

resi

den

tsrsquop

redi

cted

ran

ksin

the

dest

inat

ion

vers

us

the

orig

in

inte

ract

edw

ith

the

age

ofth

ech

ild

atth

eti

me

ofth

em

ove

(m)

We

perm

itse

para

teli

nea

rin

tera

ctio

ns

for

m

23an

dm

gt23

and

repo

rtth

eco

effi

cien

ton

the

inte

ract

ion

for

m

23E

ach

regr

essi

onal

soin

clu

des

addi

tion

alco

ntr

ols

spec

ified

ineq

uat

ion

(7)

Per

man

ent

resi

den

tsrsquop

redi

cted

ran

ksar

eco

nst

ruct

edu

sin

gli

nea

rre

gres

sion

sof

chil

dren

rsquosra

nks

onpa

ren

tsrsquor

anks

inea

chC

Zan

dbi

rth

coh

ort

assh

own

inF

igu

reI

Col

um

n(1

)re

port

sth

ees

tim

ate

ofγ

from

equ

atio

n(7

)u

sin

gal

lch

ildr

enin

the

prim

ary

anal

ysis

sam

ple

ofon

e-ti

me

mov

ers

defi

ned

inth

en

otes

toT

able

I(P

anel

C)

Col

um

ns

(2)

and

(3)

rest

rict

the

sam

ple

toth

ose

wh

om

ove

ator

befo

reag

e23

or18

In

colu

mn

(4)

we

excl

ude

the

coh

ort

inte

ract

ion

sw

ith

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

loca

tion

and

inst

ead

incl

ude

asi

ngl

eco

ntr

olfo

rth

epr

edic

ted

outc

omes

ofpe

rman

ent

resi

den

tsin

the

orig

inC

olu

mn

(5)r

epli

cate

sco

lum

n(1

)u

sin

gin

divi

dual

inco

me

ran

ks(r

ath

erth

anh

ouse

hol

din

com

era

nks

)to

mea

sure

both

the

chil

drsquos

outc

ome

and

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

Col

um

n(6

)ad

dsfi

xed

effe

cts

for

the

chil

drsquos

CZ

in20

12to

the

spec

ifica

tion

inco

lum

n(1

)n

ote

that

this

spec

ifica

tion

has

asl

igh

tly

smal

ler

nu

mbe

rof

obse

rvat

ion

sbe

cau

seZ

IPco

dein

form

atio

nis

mis

sin

gfo

rso

me

chil

dren

in20

12(e

g

ach

ild

wit

hn

oea

rnin

gsor

taxa

ble

inco

me)

Col

um

n(7

)ad

dsfa

mil

yfi

xed

effe

cts

toth

eba

seli

ne

spec

ifica

tion

inco

lum

n(1

)C

olu

mn

(8)

adds

fam

ily

fixe

def

fect

sto

the

spec

ifica

tion

inco

lum

n(4

)th

atdo

esn

otin

clu

deco

hor

t-va

ryin

gin

terc

epts

C

olu

mn

(9)

adds

con

trol

sfo

rch

ange

sin

pare

nta

lm

arit

alst

atu

san

din

com

era

nk

inth

eye

arbe

fore

vers

us

afte

rth

em

ove

alon

gw

ith

thei

rin

tera

ctio

ns

wit

hth

eag

eof

the

chil

dat

the

tim

eof

the

mov

ean

din

dica

tors

for

mov

ing

abov

ean

dbe

low

age

23t

oth

esp

ecifi

cati

onin

colu

mn

(7)





IVC Mechanisms
















IVD Summary




















FIGURE V






















Kz34








FIGURE VI

























FIGURE VII


















(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES




















































yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+30sum

m=9


s=1980













yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+1987sum

s=1980



)odps




sum1988s=1980 I(si = s)α2






TA

BL

EII

CH

ILD

HO

OD

EX

PO

SU

RE

EF

FE

CT

ES

TIM

AT

ES

Dep

ende

nt

vari

able

Ch

ildrsquo

sin

com

era

nk

atag

e24

Wit

hfa

mil

yfi

xed

effe

cts

Spe

cifi

cati

on

Poo

led

Age

23

Age

lt18

No

coh

ort

Indi

vidu

alC

hil

dC

ZB

asel

ine

No

coh

ort

Tim

e-co

ntr

ols

inco

me

FE

con

trol

sva

ryin

gco

ntr

ols

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Exp

osu

reef

fect

(γ)

004

00

040

003

70

036

004

10

031

004

40

031

004

3(0

002

)(0

002

)(0

005

)(0

002

)(0

002

)(0

002

)(0

008

)(0

005

)(0

008

)

Nu

mo

fob

s1

553

021

128

777

368

732

31

553

021

155

302

11

473

218

155

302

11

553

021

155

302

1

Not

esT

his

tabl

ere

port

ses

tim

ates

ofan

nu

alch

ildh

ood

expo

sure

effe

cts

onch

ildr

enrsquos

inco

me

ran

ksat

age

24(γ

)T

he

esti

mat

esca

nbe

inte

rpre

ted

asth

eim

pact

ofsp

endi

ng

anad

diti

onal

year

ofch

ildh

ood

ina

CZ

wh

ere

chil

dren

ofpe

rman

ent

resi

den

tsh

ave

1pe

rcen

tile

poin

th

igh

erin

com

era

nks

atag

e24

Sta

nda

rder

rors

are

show

nin

pare

nth

eses

Eac

hco

lum

nre

port

ses

tim

ates

from

are

gres

sion

ofa

chil

drsquos

inco

me

ran

kat

age

24on

the

diff

eren

cebe

twee

npe

rman

ent

resi

den

tsrsquop

redi

cted

ran

ksin

the

dest

inat

ion

vers

us

the

orig

in

inte

ract

edw

ith

the

age

ofth

ech

ild

atth

eti

me

ofth

em

ove

(m)

We

perm

itse

para

teli

nea

rin

tera

ctio

ns

for

m

23an

dm

gt23

and

repo

rtth

eco

effi

cien

ton

the

inte

ract

ion

for

m

23E

ach

regr

essi

onal

soin

clu

des

addi

tion

alco

ntr

ols

spec

ified

ineq

uat

ion

(7)

Per

man

ent

resi

den

tsrsquop

redi

cted

ran

ksar

eco

nst

ruct

edu

sin

gli

nea

rre

gres

sion

sof

chil

dren

rsquosra

nks

onpa

ren

tsrsquor

anks

inea

chC

Zan

dbi

rth

coh

ort

assh

own

inF

igu

reI

Col

um

n(1

)re

port

sth

ees

tim

ate

ofγ

from

equ

atio

n(7

)u

sin

gal

lch

ildr

enin

the

prim

ary

anal

ysis

sam

ple

ofon

e-ti

me

mov

ers

defi

ned

inth

en

otes

toT

able

I(P

anel

C)

Col

um

ns

(2)

and

(3)

rest

rict

the

sam

ple

toth

ose

wh

om

ove

ator

befo

reag

e23

or18

In

colu

mn

(4)

we

excl

ude

the

coh

ort

inte

ract

ion

sw

ith

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

loca

tion

and

inst

ead

incl

ude

asi

ngl

eco

ntr

olfo

rth

epr

edic

ted

outc

omes

ofpe

rman

ent

resi

den

tsin

the

orig

inC

olu

mn

(5)r

epli

cate

sco

lum

n(1

)u

sin

gin

divi

dual

inco

me

ran

ks(r

ath

erth

anh

ouse

hol

din

com

era

nks

)to

mea

sure

both

the

chil

drsquos

outc

ome

and

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

Col

um

n(6

)ad

dsfi

xed

effe

cts

for

the

chil

drsquos

CZ

in20

12to

the

spec

ifica

tion

inco

lum

n(1

)n

ote

that

this

spec

ifica

tion

has

asl

igh

tly

smal

ler

nu

mbe

rof

obse

rvat

ion

sbe

cau

seZ

IPco

dein

form

atio

nis

mis

sin

gfo

rso

me

chil

dren

in20

12(e

g

ach

ild

wit

hn

oea

rnin

gsor

taxa

ble

inco

me)

Col

um

n(7

)ad

dsfa

mil

yfi

xed

effe

cts

toth

eba

seli

ne

spec

ifica

tion

inco

lum

n(1

)C

olu

mn

(8)

adds

fam

ily

fixe

def

fect

sto

the

spec

ifica

tion

inco

lum

n(4

)th

atdo

esn

otin

clu

deco

hor

t-va

ryin

gin

terc

epts

C

olu

mn

(9)

adds

con

trol

sfo

rch

ange

sin

pare

nta

lm

arit

alst

atu

san

din

com

era

nk

inth

eye

arbe

fore

vers

us

afte

rth

em

ove

alon

gw

ith

thei

rin

tera

ctio

ns

wit

hth

eag

eof

the

chil

dat

the

tim

eof

the

mov

ean

din

dica

tors

for

mov

ing

abov

ean

dbe

low

age

23t

oth

esp

ecifi

cati

onin

colu

mn

(7)





IVC Mechanisms
















IVD Summary




















FIGURE V






















Kz34








FIGURE VI

























FIGURE VII


















(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES















































yi =1988sum

s=1980

I(si = s)(α1

s + α2s ypos

)+

30summ=9

I(mi = m)(ζ 1

m + ζ 2mpi

)

+1987sum

s=1980



)odps




sum1988s=1980 I(si = s)α2






TA

BL

EII

CH

ILD

HO

OD

EX

PO

SU

RE

EF

FE

CT

ES

TIM

AT

ES

Dep

ende

nt

vari

able

Ch

ildrsquo

sin

com

era

nk

atag

e24

Wit

hfa

mil

yfi

xed

effe

cts

Spe

cifi

cati

on

Poo

led

Age

23

Age

lt18

No

coh

ort

Indi

vidu

alC

hil

dC

ZB

asel

ine

No

coh

ort

Tim

e-co

ntr

ols

inco

me

FE

con

trol

sva

ryin

gco

ntr

ols

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Exp

osu

reef

fect

(γ)

004

00

040

003

70

036

004

10

031

004

40

031

004

3(0

002

)(0

002

)(0

005

)(0

002

)(0

002

)(0

002

)(0

008

)(0

005

)(0

008

)

Nu

mo

fob

s1

553

021

128

777

368

732

31

553

021

155

302

11

473

218

155

302

11

553

021

155

302

1

Not

esT

his

tabl

ere

port

ses

tim

ates

ofan

nu

alch

ildh

ood

expo

sure

effe

cts

onch

ildr

enrsquos

inco

me

ran

ksat

age

24(γ

)T

he

esti

mat

esca

nbe

inte

rpre

ted

asth

eim

pact

ofsp

endi

ng

anad

diti

onal

year

ofch

ildh

ood

ina

CZ

wh

ere

chil

dren

ofpe

rman

ent

resi

den

tsh

ave

1pe

rcen

tile

poin

th

igh

erin

com

era

nks

atag

e24

Sta

nda

rder

rors

are

show

nin

pare

nth

eses

Eac

hco

lum

nre

port

ses

tim

ates

from

are

gres

sion

ofa

chil

drsquos

inco

me

ran

kat

age

24on

the

diff

eren

cebe

twee

npe

rman

ent

resi

den

tsrsquop

redi

cted

ran

ksin

the

dest

inat

ion

vers

us

the

orig

in

inte

ract

edw

ith

the

age

ofth

ech

ild

atth

eti

me

ofth

em

ove

(m)

We

perm

itse

para

teli

nea

rin

tera

ctio

ns

for

m

23an

dm

gt23

and

repo

rtth

eco

effi

cien

ton

the

inte

ract

ion

for

m

23E

ach

regr

essi

onal

soin

clu

des

addi

tion

alco

ntr

ols

spec

ified

ineq

uat

ion

(7)

Per

man

ent

resi

den

tsrsquop

redi

cted

ran

ksar

eco

nst

ruct

edu

sin

gli

nea

rre

gres

sion

sof

chil

dren

rsquosra

nks

onpa

ren

tsrsquor

anks

inea

chC

Zan

dbi

rth

coh

ort

assh

own

inF

igu

reI

Col

um

n(1

)re

port

sth

ees

tim

ate

ofγ

from

equ

atio

n(7

)u

sin

gal

lch

ildr

enin

the

prim

ary

anal

ysis

sam

ple

ofon

e-ti

me

mov

ers

defi

ned

inth

en

otes

toT

able

I(P

anel

C)

Col

um

ns

(2)

and

(3)

rest

rict

the

sam

ple

toth

ose

wh

om

ove

ator

befo

reag

e23

or18

In

colu

mn

(4)

we

excl

ude

the

coh

ort

inte

ract

ion

sw

ith

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

loca

tion

and

inst

ead

incl

ude

asi

ngl

eco

ntr

olfo

rth

epr

edic

ted

outc

omes

ofpe

rman

ent

resi

den

tsin

the

orig

inC

olu

mn

(5)r

epli

cate

sco

lum

n(1

)u

sin

gin

divi

dual

inco

me

ran

ks(r

ath

erth

anh

ouse

hol

din

com

era

nks

)to

mea

sure

both

the

chil

drsquos

outc

ome

and

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

Col

um

n(6

)ad

dsfi

xed

effe

cts

for

the

chil

drsquos

CZ

in20

12to

the

spec

ifica

tion

inco

lum

n(1

)n

ote

that

this

spec

ifica

tion

has

asl

igh

tly

smal

ler

nu

mbe

rof

obse

rvat

ion

sbe

cau

seZ

IPco

dein

form

atio

nis

mis

sin

gfo

rso

me

chil

dren

in20

12(e

g

ach

ild

wit

hn

oea

rnin

gsor

taxa

ble

inco

me)

Col

um

n(7

)ad

dsfa

mil

yfi

xed

effe

cts

toth

eba

seli

ne

spec

ifica

tion

inco

lum

n(1

)C

olu

mn

(8)

adds

fam

ily

fixe

def

fect

sto

the

spec

ifica

tion

inco

lum

n(4

)th

atdo

esn

otin

clu

deco

hor

t-va

ryin

gin

terc

epts

C

olu

mn

(9)

adds

con

trol

sfo

rch

ange

sin

pare

nta

lm

arit

alst

atu

san

din

com

era

nk

inth

eye

arbe

fore

vers

us

afte

rth

em

ove

alon

gw

ith

thei

rin

tera

ctio

ns

wit

hth

eag

eof

the

chil

dat

the

tim

eof

the

mov

ean

din

dica

tors

for

mov

ing

abov

ean

dbe

low

age

23t

oth

esp

ecifi

cati

onin

colu

mn

(7)





IVC Mechanisms
















IVD Summary




















FIGURE V






















Kz34








FIGURE VI

























FIGURE VII


















(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES











































TA

BL

EII

CH

ILD

HO

OD

EX

PO

SU

RE

EF

FE

CT

ES

TIM

AT

ES

Dep

ende

nt

vari

able

Ch

ildrsquo

sin

com

era

nk

atag

e24

Wit

hfa

mil

yfi

xed

effe

cts

Spe

cifi

cati

on

Poo

led

Age

23

Age

lt18

No

coh

ort

Indi

vidu

alC

hil

dC

ZB

asel

ine

No

coh

ort

Tim

e-co

ntr

ols

inco

me

FE

con

trol

sva

ryin

gco

ntr

ols

(1)

(2)

(3)

(4)

(5)

(6)

(7)

(8)

(9)

Exp

osu

reef

fect

(γ)

004

00

040

003

70

036

004

10

031

004

40

031

004

3(0

002

)(0

002

)(0

005

)(0

002

)(0

002

)(0

002

)(0

008

)(0

005

)(0

008

)

Nu

mo

fob

s1

553

021

128

777

368

732

31

553

021

155

302

11

473

218

155

302

11

553

021

155

302

1

Not

esT

his

tabl

ere

port

ses

tim

ates

ofan

nu

alch

ildh

ood

expo

sure

effe

cts

onch

ildr

enrsquos

inco

me

ran

ksat

age

24(γ

)T

he

esti

mat

esca

nbe

inte

rpre

ted

asth

eim

pact

ofsp

endi

ng

anad

diti

onal

year

ofch

ildh

ood

ina

CZ

wh

ere

chil

dren

ofpe

rman

ent

resi

den

tsh

ave

1pe

rcen

tile

poin

th

igh

erin

com

era

nks

atag

e24

Sta

nda

rder

rors

are

show

nin

pare

nth

eses

Eac

hco

lum

nre

port

ses

tim

ates

from

are

gres

sion

ofa

chil

drsquos

inco

me

ran

kat

age

24on

the

diff

eren

cebe

twee

npe

rman

ent

resi

den

tsrsquop

redi

cted

ran

ksin

the

dest

inat

ion

vers

us

the

orig

in

inte

ract

edw

ith

the

age

ofth

ech

ild

atth

eti

me

ofth

em

ove

(m)

We

perm

itse

para

teli

nea

rin

tera

ctio

ns

for

m

23an

dm

gt23

and

repo

rtth

eco

effi

cien

ton

the

inte

ract

ion

for

m

23E

ach

regr

essi

onal

soin

clu

des

addi

tion

alco

ntr

ols

spec

ified

ineq

uat

ion

(7)

Per

man

ent

resi

den

tsrsquop

redi

cted

ran

ksar

eco

nst

ruct

edu

sin

gli

nea

rre

gres

sion

sof

chil

dren

rsquosra

nks

onpa

ren

tsrsquor

anks

inea

chC

Zan

dbi

rth

coh

ort

assh

own

inF

igu

reI

Col

um

n(1

)re

port

sth

ees

tim

ate

ofγ

from

equ

atio

n(7

)u

sin

gal

lch

ildr

enin

the

prim

ary

anal

ysis

sam

ple

ofon

e-ti

me

mov

ers

defi

ned

inth

en

otes

toT

able

I(P

anel

C)

Col

um

ns

(2)

and

(3)

rest

rict

the

sam

ple

toth

ose

wh

om

ove

ator

befo

reag

e23

or18

In

colu

mn

(4)

we

excl

ude

the

coh

ort

inte

ract

ion

sw

ith

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

loca

tion

and

inst

ead

incl

ude

asi

ngl

eco

ntr

olfo

rth

epr

edic

ted

outc

omes

ofpe

rman

ent

resi

den

tsin

the

orig

inC

olu

mn

(5)r

epli

cate

sco

lum

n(1

)u

sin

gin

divi

dual

inco

me

ran

ks(r

ath

erth

anh

ouse

hol

din

com

era

nks

)to

mea

sure

both

the

chil

drsquos

outc

ome

and

the

pred

icte

dou

tcom

esof

perm

anen

tre

side

nts

inth

eor

igin

and

dest

inat

ion

Col

um

n(6

)ad

dsfi

xed

effe

cts

for

the

chil

drsquos

CZ

in20

12to

the

spec

ifica

tion

inco

lum

n(1

)n

ote

that

this

spec

ifica

tion

has

asl

igh

tly

smal

ler

nu

mbe

rof

obse

rvat

ion

sbe

cau

seZ

IPco

dein

form

atio

nis

mis

sin

gfo

rso

me

chil

dren

in20

12(e

g

ach

ild

wit

hn

oea

rnin

gsor

taxa

ble

inco

me)

Col

um

n(7

)ad

dsfa

mil

yfi

xed

effe

cts

toth

eba

seli

ne

spec

ifica

tion

inco

lum

n(1

)C

olu

mn

(8)

adds

fam

ily

fixe

def

fect

sto

the

spec

ifica

tion

inco

lum

n(4

)th

atdo

esn

otin

clu

deco

hor

t-va

ryin

gin

terc

epts

C

olu

mn

(9)

adds

con

trol

sfo

rch

ange

sin

pare

nta

lm

arit

alst

atu

san

din

com

era

nk

inth

eye

arbe

fore

vers

us

afte

rth

em

ove

alon

gw

ith

thei

rin

tera

ctio

ns

wit

hth

eag

eof

the

chil

dat

the

tim

eof

the

mov

ean

din

dica

tors

for

mov

ing

abov

ean

dbe

low

age

23t

oth

esp

ecifi

cati

onin

colu

mn

(7)





IVC Mechanisms
















IVD Summary




















FIGURE V






















Kz34








FIGURE VI

























FIGURE VII


















(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES













































IVC Mechanisms
















IVD Summary




















FIGURE V






















Kz34








FIGURE VI

























FIGURE VII


















(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES



















































IVD Summary




















FIGURE V






















Kz34








FIGURE VI

























FIGURE VII


















(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES


























































FIGURE V






















Kz34








FIGURE VI

























FIGURE VII


















(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES




























































Kz34








FIGURE VI

























FIGURE VII


















(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES











































FIGURE VI

























FIGURE VII


















(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES



























































FIGURE VII


















(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES





















































(1) (2) (3) (4) (5) (6)



Num of obs 1553021 1553021 1553021 1553021 1553021 1553021








odps and












(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES




















































(9) Cov(θi m

qodps|Xq

)gt 0 rArr Cov

(θi modps|Xq m

qodps

)gt 0










(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES














































(1) (2) (3) (4) (5) (6) (7)




Num of obs 1552898 1552898 1552898 1552898 1552898 1552898 490964








godps











VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES














































VE Summary








VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES














































VI OTHER OUTCOMES















FIGURE VIII







VII CONCLUSION














REFERENCES





















































FIGURE VIII







VII CONCLUSION














REFERENCES













































VII CONCLUSION














REFERENCES



















































REFERENCES










































the quarterly journal of economics · 2018. 7. 11. · the quarterly journal of economics vol. 133...

Documents